<%BANNER%>

Effects of Spectral Slope on Perceived Breathiness in Vowels

xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20110217_AAAACW INGEST_TIME 2011-02-17T22:08:18Z PACKAGE UFE0014823_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 1051966 DFID F20110217_AACFUY ORIGIN DEPOSITOR PATH landera_m_Page_30.jp2 GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
cf3a4fe113ede64be348ed083080da1d
SHA-1
ca9789ee4e6f4e4f5cbf3e9fb3337e724dd51956
2633 F20110217_AACFNC landera_m_Page_58.txt
e757679d51e8a972536fb9bfc12224fb
49eb3040d9297187740556c6aee4cf9a8a589816
7326 F20110217_AACFIF landera_m_Page_57thm.jpg
0d9cf7e5d4c164009d62d83f00ad2061
8f6f06ba137798d0ef0db10f1ca309961c32aa4f
28108 F20110217_AACFSA landera_m_Page_35.QC.jpg
a0977078f15c54cf27f4857a2860e872
8e066d173c2e63335e011362e16fd9de6bebde76
1051967 F20110217_AACFUZ landera_m_Page_31.jp2
9c39e2298de3ccf4d26ffd6c5202c9a1
cf6fae2a0985f9164ce6f620efade103b586c188
767 F20110217_AACFND landera_m_Page_59.txt
72bd08ed5d337bec80ce0f3592215d50
fb7c7086928974559d93f015058782e3f81ecad1
215998 F20110217_AACFIG landera_m_Page_09.jp2
e015ea0b045b998efa20e981cdaa7a10
b08acd5f6be723907423c0ad4890ab5295a5f0d4
76933 F20110217_AACFSB landera_m_Page_36.jpg
c85b3f6484e7d84b198d9f75c28bb4a3
3dbc3e4ecb199987f3bae58cdf323dd08c0b34e0
1784 F20110217_AACFNE landera_m_Page_60.txt
75b571f95a64fba9957ed8f5f55b6dd6
626d5dc3ba9964cf15757d73309a67b94c7aa91d
13916 F20110217_AACFIH landera_m_Page_55.pro
488b5e4d03ac0000bcaeb9e8dda99ba9
356196060e8a503b3879c4f58a0bcb563c2b1bdd
26005 F20110217_AACFSC landera_m_Page_36.QC.jpg
07b01927b171a84e6dc3a962639813d1
de4d1679122669e6f62451bf899a0af834b269d1
690 F20110217_AACFNF landera_m_Page_61.txt
87199c20108d0ce6212fbde30a5df0c2
c9dcd14ef8a93946768ce978edd2e6871778f01d
6808 F20110217_AACFXA landera_m_Page_25thm.jpg
0e78fb977c31fe1ad8c2c6883e761094
558835a904ed150076f8c054de24a4679075d9b1
5693 F20110217_AACFII landera_m_Page_56thm.jpg
5a48960b7ab992bdd8d0c2c0a5e30532
db5c566cf20aa705b62c9b17c17297726b71c33f
68307 F20110217_AACFSD landera_m_Page_37.jpg
ff2db7a7622b40c62eeaa4397389ed0e
f00dd61b75ddd6917d8b35b6ef2857c95bc9ce3a
7393 F20110217_AACFNG landera_m_Page_01.pro
613e725d5891026d25e8ab7ea0e5bb75
4b9353c77522674e772a99a04d9e97e8cfa6cada
8423998 F20110217_AACFIJ landera_m_Page_34.tif
4d41d0331df730d8945e89346118d48b
9e39264d450ffcc94ece49b576b31c3408e8fe81
21572 F20110217_AACFSE landera_m_Page_37.QC.jpg
60ada288f0c1fb156c65d0e531822718
332a346ea1e6bbf608b7216457e739ed411439a0
1331 F20110217_AACFNH landera_m_Page_02.pro
58e7e4df1a1df7e44a3349f6a312f39d
7657a295428b6cd733ffebcb228669a75709ce41
8371 F20110217_AACFXB landera_m_Page_26thm.jpg
e2da49caf7ab91836c04c3953444c622
7e310edcb820576c79659276e8794b4d40b4e05b
49116 F20110217_AACFIK landera_m_Page_16.pro
a8fcf3fdf90c490c592a1c7baed3a81b
57dd64cf6f5629d4e97b7d8e0a0b99e3e75a60b4
8553 F20110217_AACFXC landera_m_Page_27thm.jpg
3f63ae9046545e0e7ab0d3f24176ba7e
129d1f35f6edc3d8a294bdf6f394ecf047e69742
82915 F20110217_AACFSF landera_m_Page_38.jpg
1bbb22a6cbac643ecf83f938afd5c95e
e34636ff7d08badfdfe997dc6ac9cf793e833343
39850 F20110217_AACFNI landera_m_Page_03.pro
a0756f6f4650f73777b948789b8dc356
6cc00da336740bd29a6044f47bc82c68e17344c1
7094 F20110217_AACFIL landera_m_Page_38thm.jpg
d28eda70380796b2164149901c6f0454
f6abeb9cdbea0d2c561133f09fec1b4fa8aa47ab
8684 F20110217_AACFXD landera_m_Page_28thm.jpg
998184d75735b876c158b758fcfe4f77
79c888ed0ea63ca77da118bb9c342486ed26fc16
26582 F20110217_AACFSG landera_m_Page_38.QC.jpg
49603ee81d52207ddabc3f82352d9759
122dd305d173fedea3573933d1d5cf0c2441165d
71648 F20110217_AACFNJ landera_m_Page_04.pro
c610a84d90de717194e8457c83dc8ed9
257f7ea6571d3a8f89c9f42eba56c6aaad984f9f
33006 F20110217_AACFIM landera_m_Page_54.QC.jpg
9d61d9eea153c4270a2397d4e97cc43b
d75a5e32ca83b07e5d7748598438a665eb61b32a
8084 F20110217_AACFXE landera_m_Page_29thm.jpg
10386d0959f4c883da8a36ba8605ca54
74635a02260a09f42b735de0326e9df11a70598c
32409 F20110217_AACFSH landera_m_Page_39.jpg
362bbf82b7819e1f24401572e6ebd682
61419bccd3578d800be0d701b28833e31f6726f9
11164 F20110217_AACFNK landera_m_Page_05.pro
6693400a3039bbfa39778b11db94ff06
e4ab14b6bf9361e948580843e4ab6627529fed32
654968 F20110217_AACFIN landera_m_Page_04.jp2
f7d66c9a86e09ce2c6bdf426467f52e7
d9750ac9eff87854c45fdf907823bccc3acee97a
8095 F20110217_AACFXF landera_m_Page_30thm.jpg
f305959e11b1c51fdac40ba3cdf1cabb
dad2f15e8f0402c31dedaaef6217ef706a4d05d6
9980 F20110217_AACFSI landera_m_Page_39.QC.jpg
23f40f251bdd3437e015d6ef331ca8ac
0656cf7c70ade258085615f0b7e7c1e430c92514
25182 F20110217_AACFNL landera_m_Page_06.pro
ec7211ad054eff79a1327ed8b9ef1488
33684f5c5510363a252affca8128b3aab211dc61
105006 F20110217_AACFIO landera_m_Page_28.jpg
07db428a4e0b3bc0fd89fcc46e039770
e66f6c04ed376fa083363c85c4a095154b581e3c
8363 F20110217_AACFXG landera_m_Page_31thm.jpg
29a80d8fc449055267c6231b34e549be
8cd9fc4d09b763531787c858e4eae88c2197a145
101219 F20110217_AACFSJ landera_m_Page_40.jpg
0180b36342944aa8a84f849913700fe2
f996894627057fd8a5d8d4dccde5af82b72bf002
23885 F20110217_AACFNM landera_m_Page_07.pro
9024a0d3f37e83cd6998e649600f340a
5db244ef3dcd26096a17342264502cb215c23b29
5062 F20110217_AACFIP landera_m_Page_02.jpg
eb457cf67098f255436b58406eba7e57
c043ffd2313a0ff9545e30756819c160b2e0042b
6346 F20110217_AACFXH landera_m_Page_32thm.jpg
2f9387dc3e77ae4cba2483c3bb4e7511
b02ea103c145bba79e87bf1c6c8b66c81653981d
41736 F20110217_AACFNN landera_m_Page_08.pro
67dda255c812bec008665c19f609368d
48df536af5e3c3402a31002693bb3aaab37ec6a2
100671 F20110217_AACFIQ UFE0014823_00001.xml FULL
41fc26b242773edd1197d04fc3481aca
b53573cfa5d3c2e98ca07c7118c70029ab2e063e
34295 F20110217_AACFSK landera_m_Page_40.QC.jpg
a38aa2543a0cb8dbda8ca8af98da1db1
2703c913d42f143d835ac4e95a60777037a5c996
5415 F20110217_AACFXI landera_m_Page_34thm.jpg
f06a6e681fb7080aae5815e557a5e200
727c07d12977542e107cc9387cc31b1d84d1422c
9785 F20110217_AACFNO landera_m_Page_09.pro
bd2314d7480ca0bc4bd53d034141acf9
0b6fe56e07f5e4789727904d6743d693d2b49598
98299 F20110217_AACFSL landera_m_Page_41.jpg
11ed5ecfc835bae7b4843eed2ff0ebd5
8878b65e060bbf62bc4bbf984c6b6529c1011fd4
7428 F20110217_AACFXJ landera_m_Page_35thm.jpg
bf148cfbf2180d62dd75560598eb4c0e
40e2844d4f30341b333273b8ea46014b0061d2a8
44140 F20110217_AACFNP landera_m_Page_10.pro
855f8454c1446aa452a0aa27ea2f14b7
f9c704e346383e78d089b6e3e7f6f90361b151ff
31747 F20110217_AACFSM landera_m_Page_41.QC.jpg
0b7a0246f3abdcd58ae3128b0b8779e1
15e47c8a17d663ec931b86b289b02b337db2d987
6084 F20110217_AACFXK landera_m_Page_36thm.jpg
4a892bc388cdafada2c6ca21f194dcc1
e89321ebd40071ce559f4753afe4d0a669c42d70
50682 F20110217_AACFNQ landera_m_Page_11.pro
43db9ae45344aeba2ed2c643779285b7
d1c5cbf63a2b1f8a18f906939c715d14893f8190
27982 F20110217_AACFSN landera_m_Page_42.QC.jpg
d161af913affa84d639af1228c157c3e
1d33bf25f7dfe76312f8640bc5b450a22468313e
5256 F20110217_AACFXL landera_m_Page_37thm.jpg
b46629ab4304d91cf0ecf9725d4212b7
3a36b42c406d540412150cd72fd44a6a4db8ede7
50355 F20110217_AACFNR landera_m_Page_12.pro
8e847fcc2d4fa5e498116e293cbafd2f
a3b6eba9c3cd7d2de49a6ecddc7eda82da798fed
F20110217_AACFIT landera_m_Page_01.tif
af1c00b187e21b45ad3ea420e4244e82
3452098e50e61f2e1429016a1df43303f12320ec
75664 F20110217_AACFSO landera_m_Page_43.jpg
e9965476de5002577144ae3933b6d1b3
f07e155822321c90373d97f5c7f682bb048929d9
3332 F20110217_AACFXM landera_m_Page_39thm.jpg
e16945db8d3b6a2b12cc9c6d18787ba3
53a7a45393a99d55b6aefbe2979284a19fa6afd6
46481 F20110217_AACFNS landera_m_Page_13.pro
4b7c4be35d83c342d54c397a907427e0
6564ef4f52ea7d9c999820376a6ce38f2a34e18e
F20110217_AACFIU landera_m_Page_02.tif
2c57a604f2bba7b6165130c54abcceb7
f0576a6056761baea9490e41a6b1df3b39249d6a
25126 F20110217_AACFSP landera_m_Page_43.QC.jpg
bba7cc9a7fc73659557457cdf2f68e4d
9a89c912e147d6b78e46b2925e02a279395ef891
8038 F20110217_AACFXN landera_m_Page_40thm.jpg
9f41288958459134c5ff6951d0fcb3a2
00dfead932ca6c134d8ce8a8c34170ba94520ec5
34235 F20110217_AACFNT landera_m_Page_14.pro
d6f5c71fac1141b2ccb41fadc86b406b
2226b65671ddaa536e0595bbca1434ff49fbc350
F20110217_AACFIV landera_m_Page_03.tif
f46852c20e44b6d69582d2717de33357
deb3c7c1edcbb5333d096985d14cdf9ec82e3f93
86593 F20110217_AACFSQ landera_m_Page_44.jpg
db83efa2df57867441d0f27b5448aec8
646d7eaf2733c23506c8d03c6572f37e0ac7813f
8036 F20110217_AACFXO landera_m_Page_41thm.jpg
947be23e22bae1ef884ba2cf83dadd5a
fd451ee3e5e6ca6a13ad60af53b0e9d8eb3cd305
42628 F20110217_AACFNU landera_m_Page_15.pro
68293e4590d141c7a916d97252245230
70afe35bdcc73fa03e2d27e192f9bff9d922b8e3
F20110217_AACFIW landera_m_Page_04.tif
1a106e4b909e24e5a47347c39a8988ac
630ebb034e068f39cc4deff72c4c3c27e8f85807
27981 F20110217_AACFSR landera_m_Page_44.QC.jpg
9e9f59cfc85a7ce8b9c8d9c2c09ee635
c52693fe6df7f49f0483b5c1e99e963b883bacdc
6937 F20110217_AACFXP landera_m_Page_42thm.jpg
2967e660ae712cf43c450c28b7802e05
adbc6ef6d09c87aa90d4bee1efdccac9180b43af
50184 F20110217_AACFNV landera_m_Page_17.pro
b370812a885526b7971a19ea8fb773bf
533cf1fee679edc3bd0be6fe35430ba853c1a126
F20110217_AACFIX landera_m_Page_05.tif
63773a47cf4eeb92d946bfe66211872b
ab6c1fa9b1cf65df90d9f48564074711368a2709
90464 F20110217_AACFSS landera_m_Page_45.jpg
50577c4ee5d6115b92b24c784a64f8ea
681dc019e91bc0f7efdcca41e351c061245f058e
6227 F20110217_AACFXQ landera_m_Page_43thm.jpg
6d940061b2275c9fef8b05363cf86472
75ab92c64c3b7fea5db60b3f13de6f9832b5f35d
49783 F20110217_AACFNW landera_m_Page_18.pro
cb608c428f01ca5b2344dd40e37899a8
1e9c419fd17a0ff71dd292731cab91730e23c3ac
F20110217_AACFIY landera_m_Page_06.tif
6d059c39f69850dd12516a62b65f02ab
a5119d2c164cbd701e0caa62b4ca5b3d61233f0b
29583 F20110217_AACFST landera_m_Page_45.QC.jpg
f15337eb21d88895f0bc11e450ad4446
414b0bcb39fb7ba4b9b4508ff1e4ef5ae724d04b
7330 F20110217_AACFXR landera_m_Page_44thm.jpg
d9fcc41f5f4cd1d2d1de942b147ad8ff
dce6a9eb992fb709c0f1ab11459feecaeb1285ad
F20110217_AACFIZ landera_m_Page_07.tif
6ed27a334d3dbafa5792f92d0ab0bf6e
9086aad4629ff29c026264b44b3a97ea9b857ec6
104662 F20110217_AACFSU landera_m_Page_46.jpg
2ba6cdb58ea6a3a286cdd98726fc77d1
286e7a77a62151c0c59582ed9f2aa4fd770294f8
7262 F20110217_AACFXS landera_m_Page_45thm.jpg
a5ea0e24428adc87bc69a22ba467c5e0
cea0ea28d2c4b4f05791322e8c7d039363f5726d
47469 F20110217_AACFNX landera_m_Page_19.pro
3dc333cf2b3209778d2e363e6592b0c9
f02b57905c9214b67e18e14dc8051355a641866c
33703 F20110217_AACFSV landera_m_Page_46.QC.jpg
b38c7ad92fade6cf18181fa5162ff523
dc17d6146f79711bea5aabddd391442033fb663d
8539 F20110217_AACFXT landera_m_Page_46thm.jpg
3ea161d528003038dc964533dea0b954
857d071e43d0fefae61515e53e8a6a5d81505af8
51383 F20110217_AACFNY landera_m_Page_20.pro
bdf7c5fd573ec702abe8665e68b724bb
7e4a4c415eb919d51fe7be91ab7a84aafffc50d3
104396 F20110217_AACFSW landera_m_Page_47.jpg
dde87e1db505b6a009734270e982991f
9aed0bf17c410686c1bc49abc6ba73248767acb6
409 F20110217_AACFLA landera_m_Page_01.txt
4beb9fe0cf16312ecbca536a22e1d136
e85dfd958f1eaffcf914a8491db242fa9b64cb53
8683 F20110217_AACFXU landera_m_Page_47thm.jpg
814085ce76a2a5dcadf700980773148a
38ea3299dc5f2e2ca00a1d7cb0e4789d5f1c48bd
48396 F20110217_AACFNZ landera_m_Page_21.pro
df8d5eb75664e9f1f0260a61a2a23da9
b6d6b6edd70abd32f88275d25addde4ad1285231
33838 F20110217_AACFSX landera_m_Page_47.QC.jpg
870fc4020143c636e43878e81b323c43
628a3abb13f41609cf776998933919c80c25c11c
122 F20110217_AACFLB landera_m_Page_02.txt
d4949e9cfc5b2bba1ee666a8f7854c96
a9248c16e2d0010431be39e1a11f1255402ef6cf
8459 F20110217_AACFXV landera_m_Page_49thm.jpg
cb221b52e21972ecf800ca11eb629a2d
97768fc274c01d1d85382987a2b46d549624942d
99608 F20110217_AACFSY landera_m_Page_48.jpg
04b6736d511a9c8c126e15744f932e51
7d1fddfbc09e4b9c08236c9721831705089be741
1639 F20110217_AACFLC landera_m_Page_03.txt
eb2e04d8faa7211704492a8e5e3480ab
33cc363ceb71bf12862378abdadf196bdfa13155
8275 F20110217_AACFXW landera_m_Page_51thm.jpg
a822f0a612db9ad9e50b544d633d9eb5
01b61236dd3117e3126c64030ca0b7bcef048633
27508 F20110217_AACFQA landera_m_Page_08.QC.jpg
413ad03fa2e5777b85e42e84a869a747
77c7d7cdc10370e90fd0fb7d37927cfa00c5dd7a
32455 F20110217_AACFSZ landera_m_Page_48.QC.jpg
9d866f9036588302fb6fa2fd0ed041c1
3ee021d0d9de59d1d6ae9a138948cde2ca2faeec
2925 F20110217_AACFLD landera_m_Page_04.txt
38c881a0311bae4cd20cebd27da9211e
2545f1904b773c61e0aef1c47aec3fb3f7b1c469
7471 F20110217_AACFXX landera_m_Page_52thm.jpg
693d9c1c52217286141c3f1daaaa2a4d
23f896154f0f53d11a1e9480d1ecbe675ff9f0fc
22589 F20110217_AACFQB landera_m_Page_09.jpg
1317c9cf76dc24e8716db376e3a9bd83
deb4a6c1be6a9860f373f406bf791bf6b57558ed
583 F20110217_AACFLE landera_m_Page_05.txt
2e1f428a1b1895ee9ab867839cf4a862
872960b1a68cb1d6ac28e142d6e904ddedc78a81
7136 F20110217_AACFXY landera_m_Page_53thm.jpg
be139fb3f04240b1d01ff0f50af3b0d0
3be9f05846168b2aabb9af8df1a6b233461ad2f0
7342 F20110217_AACFQC landera_m_Page_09.QC.jpg
fa882a65b9f735b952ac4792ab65e182
a8a59ce02c0fe73e902f880af08b41d538c39dbf
1101 F20110217_AACFLF landera_m_Page_06.txt
f592221b89e00ce99a15ac31c00d4ffb
949b5e400af2b8510870e307083030cd7592e33f
8299 F20110217_AACFXZ landera_m_Page_54thm.jpg
47b7f58e959b5b4af96ac4ab68cd232d
ef797f111a5e6f1fb87b0dbfea6d6123fce962b5
784515 F20110217_AACFVA landera_m_Page_32.jp2
17e1b71518e759206c55445bc8c90d05
87155325c45c4153ef88b55bf9c0aea097ea30c5
89707 F20110217_AACFQD landera_m_Page_10.jpg
bd4f3f135c2f9bea130c8a397e287865
9fe5303cab5db5e5b7fcc4427448ea89523b8701
972 F20110217_AACFLG landera_m_Page_07.txt
e0b71a7f505883f1f5ee78525867ca63
a2642d9863e6713a47d3d9982e48728a82e99470
545788 F20110217_AACFVB landera_m_Page_33.jp2
db7e2f90ba61352dbe95319e0b63d220
2b21757d9f38419a11f9666328be261a1e3dd8d7
28932 F20110217_AACFQE landera_m_Page_10.QC.jpg
be5657f41b968476808400bd3e79bdbc
30db9c30217da3588064a6d00a44e98feeca555a
1805 F20110217_AACFLH landera_m_Page_08.txt
11bbf0101df6e62acc22b0e033ed418e
3e3e42a25140369adee46c9fb518483e6d5d5e7f
707400 F20110217_AACFVC landera_m_Page_34.jp2
1ab6ba53482adaa3a34a6a98e64dc68e
30b8ed8b009f9e602bfe4237419f8d36e43c0674
101319 F20110217_AACFQF landera_m_Page_11.jpg
f917dc57300d90d43a5b3df37ff85b97
fca91853e0f6edd0f607962784c1f79c1099d213
392 F20110217_AACFLI landera_m_Page_09.txt
9f89290cd77e256173124f1ae93e3831
68d60b72bb1ef2b123fadfae75180232b06fa5bb
881889 F20110217_AACFVD landera_m_Page_35.jp2
82c3601720528a1775b16c33c2ff5170
8e08c23695f881cbd67d055ede4da310466e5f70
32518 F20110217_AACFQG landera_m_Page_11.QC.jpg
4ddbdaf45da8acf43d11f3bf2caa4583
87636a27109c252b7ac8413aff20a513f12cf844
1817 F20110217_AACFLJ landera_m_Page_10.txt
faa4e7e9a6c2eeaa06665b38dff216b3
6e6a46e1713483b049a174b1b064ea274dfdce9e
763368 F20110217_AACFVE landera_m_Page_36.jp2
723af93a96bffbe680de76171af922c5
bf09a63c84c918fd9cfc593883e09ca966b06079
101214 F20110217_AACFQH landera_m_Page_12.jpg
d1f79f552305d8fdc9b3f5b08a99ce8d
fe5e7da91b9a8d0320bd9da91323e51446b977a7
2003 F20110217_AACFLK landera_m_Page_11.txt
ee544e8bf1fc383006cdbe4d61e9c1bc
eae291e71e14c54749b11feb6c1ac04c30861f1c
843774 F20110217_AACFVF landera_m_Page_38.jp2
2a79a0072fdbf94b5e82002e7d13e196
dc490a0b3f94245256f72cb2cf5c08b697341ecd
33086 F20110217_AACFQI landera_m_Page_12.QC.jpg
8fd7e9e2303ee81781dc88665a1a919e
d4eb9faf1d5dd911a20169eab4a3e5d67a4bf94e
1975 F20110217_AACFLL landera_m_Page_12.txt
12ded0df9ef60ed900781dd14f0cf695
38ca2f96d9ebc7bcbeaf4fa1ae822e24969b9157
92989 F20110217_AACFQJ landera_m_Page_13.jpg
1c08d6bbb2394db95245d82979e701e3
d26babb4fe5d8ff2e4e054fc5c1b183b5ce20eb0
1841 F20110217_AACFLM landera_m_Page_13.txt
43664761f9e872ced0a932bb1bb97b0e
6e9e3d427f52c9742da17f9848b7e1f23b97e94f
297489 F20110217_AACFVG landera_m_Page_39.jp2
65400971bdec53b99ff0f89a4cd2ccf5
ec7764f90bb7b32f086433cda0ba90b332b84fb9
29798 F20110217_AACFQK landera_m_Page_13.QC.jpg
6b58b0b603bbbc835c35d7eccf0d4284
634c2a945d2fe1e86ec06dfd09b1614f7d520190
1371 F20110217_AACFLN landera_m_Page_14.txt
4f89742664bdad595e3b93cf5a6426d3
18ae21a8c975503ffe481ea9afb74276bbd1b1ec
1051936 F20110217_AACFVH landera_m_Page_40.jp2
d7be904ce9d9a86a16b2cd1051a21c1a
079783fbb5cb28159f08f138a0cf3d1b22d6312c
70533 F20110217_AACFQL landera_m_Page_14.jpg
362df8db37f7058f0980ea7ff2055bf6
1ca3f43be4bd93e282178b1d09eb91557ce9b3a6
1802 F20110217_AACFLO landera_m_Page_15.txt
4f16883ba749265e8a86c8138247ebb8
1d1768227c2ecafb9fae2c0c187f4a818d6a843e
1023912 F20110217_AACFVI landera_m_Page_41.jp2
cfe8b14d9de9a496713800df8eabe19e
85fbfeed25121b6a9743e90a33ce3bf4d4d8864c
23503 F20110217_AACFQM landera_m_Page_14.QC.jpg
8f4ae3c4e2c1a92a88ccd77ea3c5d0e9
4c8147561fcae75ffc679c7f3a8605afd99bfa03
1946 F20110217_AACFLP landera_m_Page_16.txt
470e322ae5b1cb2a1763a2654f87b449
56d26d84ccc26c818f27211f61785e2bb6a17981
926170 F20110217_AACFVJ landera_m_Page_42.jp2
6fcd8dddadf6e171b169a0129b12699e
0e88eea3ec1ab5411ba8f0bcb0c0d8742bcc7153
86031 F20110217_AACFQN landera_m_Page_15.jpg
40f7f9d463661dab2a4d5d8453857797
cab33fe169989bea11a53150fdff84f801d65f51
2004 F20110217_AACFLQ landera_m_Page_17.txt
8a8e17a4a9b5105f2b6cc11bbcf62d9b
1768aed09aeeb8666d90679b428c7b81216bfb02
788336 F20110217_AACFVK landera_m_Page_43.jp2
a19acaf545243e9ce3078472d1de9951
84f4614853dba7b5b95f415278b44990c14c513e
27470 F20110217_AACFQO landera_m_Page_15.QC.jpg
76ed6b6fb6f575e9198d86e0b5c2a574
3d65d1f6ead2bad848f7561b78c184ecd5cc2087
1962 F20110217_AACFLR landera_m_Page_18.txt
3541362e2dd17b88d141707459d4d6a4
cc4ef6b0d963b2b8c3d671971adf4f39c028150f
987882 F20110217_AACFVL landera_m_Page_45.jp2
57fb558437378e74bf2d19eb6d244d1d
9a042e7a82ea517f1043d39512122112eaba1012
98899 F20110217_AACFQP landera_m_Page_16.jpg
3283babf3108905c8e73c64d80e741c8
007201ad176427ff4cab3711e54252825989d06b
1882 F20110217_AACFLS landera_m_Page_19.txt
5ffa6fb8f15a3c75a7f0423e326ab16b
1d2191b07b70573de010fc8f46dd004b9613472b
1051874 F20110217_AACFVM landera_m_Page_46.jp2
88f7945e62048fcebdcad9d19bb65b66
7aad41012bc65ea60c08408f93a525cec13f1cc1
31324 F20110217_AACFQQ landera_m_Page_16.QC.jpg
656e55bf9f68c72d5594aa5c7146b7e9
5b902b575f8149582ed20c51a0f33b0083ecb0cf
2060 F20110217_AACFLT landera_m_Page_20.txt
e8d0458eb7dd0df6696e8e157e8adec9
e4f2a040f11f076ed2c1b603d94200b7a708df65
1051983 F20110217_AACFVN landera_m_Page_47.jp2
dd31fd43ec8f31ff3bfd29def7de4376
ceeb85bead147543b0bd93059e94f78e250f0124
103236 F20110217_AACFQR landera_m_Page_17.jpg
bbb9bf501f5af024e68a099bb441522b
b28b6388a6008b90ab96b2c101bc6b58a12401d0
1944 F20110217_AACFLU landera_m_Page_21.txt
02437fe05484bbd80e35b79f0e1dadc7
d35b0356836201c2439c941cedc70e830aa4a0ec
1051982 F20110217_AACFVO landera_m_Page_48.jp2
ac6b8e3877ab3943b84aab60edfbfb71
60e738d5b56ab1c626c1ce2b482c9ac3b03e7d2d
34075 F20110217_AACFQS landera_m_Page_17.QC.jpg
b5d5e3f0c56dbd1ce4efe04a1f2387c2
b68bc1f68e3974a7ffd5e9ba710441966d37e421
1051919 F20110217_AACFVP landera_m_Page_49.jp2
b89095cbcdfb81de61ab6c2a22a0ce2f
13b79201e9349c47811678cf67b5da6fa51d80cb
101645 F20110217_AACFQT landera_m_Page_18.jpg
d1cfa2db28704ecd3349727bae5e38d9
cde841136ee92aa2d0a3638a24e92385c00a1c71
1992 F20110217_AACFLV landera_m_Page_22.txt
55c84023548a3f1849006dfb07e0c3d6
06f2eae4cd2451849be79bb96526c50fe712d474
1051916 F20110217_AACFVQ landera_m_Page_50.jp2
73e07c21eaddba73d91d14d4e84c8e4e
8f43d445437ecaccb6476edcc9facae0a3ad6e85
33398 F20110217_AACFQU landera_m_Page_18.QC.jpg
85b93598f814f6ea82c5e328ae270b90
5f2cb72eeb1487f32239bd90e67f4c02f31286cb
2032 F20110217_AACFLW landera_m_Page_23.txt
0a5a72b95af06857361fc49eacc6c437
0426d08fd44079402d18cbf889b65e6ad07eb131
1051965 F20110217_AACFVR landera_m_Page_51.jp2
838b53a54bddffddc80bbd49634b79ab
4f04cc80f6392df21f6ebce4bc8fc279b22254c4
96473 F20110217_AACFQV landera_m_Page_19.jpg
9d38dff6a26db12179a29a563100142f
2a82e32512bca097d8599ce3856ee3a8e6ea6f83
1807 F20110217_AACFLX landera_m_Page_24.txt
7d82ca078799193a1af571548c48da0d
8e82bedc8dada3ed529ed11a302130505d07e5a7
995897 F20110217_AACFVS landera_m_Page_52.jp2
efb0359d0244b84376c38b71b82e19fb
c55e7154265d99d39ffeb36c47abd420793c8aed
31769 F20110217_AACFQW landera_m_Page_19.QC.jpg
d639663ae3d68c1dbc21ff11ff2d48eb
2d676007a41cf2731dffb155eb8ae7a68b7068c9
F20110217_AACFJA landera_m_Page_08.tif
630baa54e8a777d0ebd171edc4166627
9abc59f69b80226d233d2fe5c916e0d51b0658e6
1759 F20110217_AACFLY landera_m_Page_25.txt
5d3130ace767a4399e091fcd511ed2f2
4a2051f68a777f2c8ac3c90f9be1408d55557182
940198 F20110217_AACFVT landera_m_Page_53.jp2
708bc0d4eb2e491a99cdb164893b76b6
bcabd75d03853c10f52dd1e12f081690ab96047f
103311 F20110217_AACFQX landera_m_Page_20.jpg
4608e5aec400ecd0343080ed28ba0cb0
ba7272c3cc56e063ea8ea15b321565afeb29b0e0
F20110217_AACFJB landera_m_Page_09.tif
1eb790485ee355a2cec0ee789fd9a156
e285f97da82eedd4ab3d5fe1f87bc9a247d778f4
2400 F20110217_AACFLZ landera_m_Page_26.txt
c25792e937f9f5b3bd9aaab673660f6c
46b27ec3a641a6f0d142bbf1bed63d15287beef9
1051975 F20110217_AACFVU landera_m_Page_54.jp2
9489068e9b9b76a1e6e386773ed4fb40
10fa4107473cc058a3e4a898f60fe436d783a897
33753 F20110217_AACFQY landera_m_Page_20.QC.jpg
4216a061111e22db75be82dd97b83510
f4b8cef52e5b17590e9e306361a70d1b2a04d6b1
F20110217_AACFJC landera_m_Page_10.tif
4eee54f0c0afe467ac786c3962ff6dcd
8da3b39a9429984df6d149ffc84a040f87341a1f
311112 F20110217_AACFVV landera_m_Page_55.jp2
6e0cc48a7d3d6efebc8d4ba5dd7c7dd5
9b321e08ab12cf32a10894c9ffcddfafdd5c8923
97548 F20110217_AACFQZ landera_m_Page_21.jpg
f22dc053dba98ad44d8eda35b62422d8
b5e12585dfba15806d378ea116d9a669d8bb7e09
F20110217_AACFJD landera_m_Page_11.tif
5951a7f588d6b1548fd9c5d83c3619d1
d32608f0ed82e703264583b6c69c19426feaf24d
851937 F20110217_AACFVW landera_m_Page_56.jp2
fb559bf8dec32bf6ee8cd0685788f117
270233f6786414e6c00e6372a633ebc33aa4a39b
49914 F20110217_AACFOA landera_m_Page_22.pro
18f0ab1417df76cbe9aee15d0b11a2af
62bf90a2aebdcbbebcb66268b1139e5ec4fac695
F20110217_AACFJE landera_m_Page_12.tif
d7b2872ed884ca72e974ce13928596a0
4d3cc0c386c2cc82523bab066854097072d933ff
1051941 F20110217_AACFVX landera_m_Page_57.jp2
bcf3779494b15cb076d065ab90a121f2
79e37b54ff8531f58448a48a78ec2da0fe923d4a
49477 F20110217_AACFOB landera_m_Page_23.pro
b384e812a1aff0b8357268bb515d7f48
f939be488ce204dad77f862f78a31c38b4ae021b
F20110217_AACFJF landera_m_Page_13.tif
d2469278f0089c3cd8652f965c09d96a
fb664a9d342342b249623df2f7c0c47951d50291
F20110217_AACFVY landera_m_Page_58.jp2
7495bc1fce3b79e2ad65ec17e58ff94a
c69962e27d5d39bfb68b592d8a11641d17424924
45480 F20110217_AACFOC landera_m_Page_24.pro
9cf491bcd74c8ddd723474851ac52f28
7ca580ec331b38e5c668a0dcb6b7bafdd00236a0
F20110217_AACFJG landera_m_Page_14.tif
7ea8dc6d577f300640cd6f46ac40e1d0
eb093b93e8e117ccc16513bd6228fc1bf3cac366
102495 F20110217_AACFTA landera_m_Page_49.jpg
fe8d3d3d7a06f050ab6989197f2737e2
c31e9bcf6efcb318a3e29f7fdf8ab105d3ef9da1
406994 F20110217_AACFVZ landera_m_Page_59.jp2
7bcde285e3d7c4bd4d0bf9fe8d9d0f66
b946c0a59094028c5ec6d63f8cdfc2b017d2b308
40518 F20110217_AACFOD landera_m_Page_25.pro
02c665de79d644233708988bfa645b9e
ef868b0a3a8d5b78cae9aed7335334c05cc72cee
F20110217_AACFJH landera_m_Page_15.tif
fc24b179f5755870566a9a121d737108
d877d673a466b81e9d03c92eb675a6fbe94e94b5
101494 F20110217_AACFTB landera_m_Page_50.jpg
bde38d7807579b27f7acc59973910cb2
2e6389515f819fdcf9552b56e7042981f58ff79a
57142 F20110217_AACFOE landera_m_Page_26.pro
de91bb9d734e9daaf7397bb737e4dfaf
4b172302eae6c1a2bd7fed50230eb0c9612ebada
8741 F20110217_AACFYA landera_m_Page_58thm.jpg
6f7b602737c9fc805b3a52784ef5f157
342c5eeeec1cc216572194f1da233abce9b6f568
F20110217_AACFJI landera_m_Page_16.tif
9c737109d05d1a1400446c6f35479e8c
3ad17be3573227c367da21e6040288e819d62f7d
33457 F20110217_AACFTC landera_m_Page_50.QC.jpg
6c7fa6be119d9210b53fc8e5b4ad0df6
e7402cb90c1f8979c40a5e564d1a3f3f4dbd28a3
52741 F20110217_AACFOF landera_m_Page_27.pro
29d5a0829360cc970db92e18b9eb5f0e
4acad4f3056448f08dfe22f5b66abbf3af23ba5f
2927 F20110217_AACFYB landera_m_Page_59thm.jpg
cf269344273726a539389e1707b858ec
8aa0d25e9b4fe36e30f9c3e37e74def63ace0101
F20110217_AACFJJ landera_m_Page_17.tif
5894fb030e8856f6eaf0b8edaf3098e5
17869c58846a7b0b7e0ba25c626fa08be9e791de
101257 F20110217_AACFTD landera_m_Page_51.jpg
1bbcf47454481a3c3df4c1c2f71517fb
41faa42972d494e9087fcf66494653c0c8be0bf0
52098 F20110217_AACFOG landera_m_Page_28.pro
16a8378b70715fe347195266352016f0
38de0eaed76783ef115a75ec58d6be58cda5cb6a
F20110217_AACFJK landera_m_Page_18.tif
7a97d4cc0ee03f78e55eb0331895b29c
a552176a1d6225f86b7c76050a03813b67440018
33159 F20110217_AACFTE landera_m_Page_51.QC.jpg
9448d8544f13bafdf3e997b1032477de
93dd50d03fca5e09a660790f7cc32f9aa9486632
49533 F20110217_AACFOH landera_m_Page_29.pro
d8aa166851dfcfbc1747599f996f30b4
c613174276637df494ef84b598d4185ba1f8a879
7394 F20110217_AACFYC landera_m_Page_60thm.jpg
cc2fb0adc3c36adf92178b6ef8c87c35
6b78586bc056636b787e61cbf263e6ea200f43fb
F20110217_AACFJL landera_m_Page_19.tif
57fef6441aad35fa78a020397e5306e8
070b4f107bcce0c0a8ef55f233495baf34ba9c87
90785 F20110217_AACFTF landera_m_Page_52.jpg
71eea0afabdbfe756e531c36d07d868c
a04cb084264ddbfa3f289e970b93bda65fd48909
48319 F20110217_AACFOI landera_m_Page_30.pro
67eb6b7780f9e2469efc55c4a856df7c
b94fd9a64ee05c1a978c89550f22f18c3ed8d520
3046 F20110217_AACFYD landera_m_Page_61thm.jpg
b65305605bd2478177cb7e5bbd73f972
31c1b58a94936fc6d9482e6f8173abc89a1a1667
F20110217_AACFJM landera_m_Page_20.tif
c6c328b7a762d1e45aeb35432de7eb69
265afe284feadc381d3834966065306e2965c86c
30255 F20110217_AACFTG landera_m_Page_52.QC.jpg
169ff8afe978fccb3250e95965047157
f442953106ad9cf63db922603be15c0442967630
51121 F20110217_AACFOJ landera_m_Page_31.pro
35841fb3ace4bc8c0011f02b56b4f06f
28aec1b846a5d2f09cb78a7376ae03f340c0f367
365628 F20110217_AACFYE landera_m.pdf
b0412c9f34441ae43df5e1973120e179
b11446edc17490e257adf32d55cbbb54c43b41de
F20110217_AACFJN landera_m_Page_21.tif
2a443f9a7c31cd7274d915620eb6b71d
851b0ed1684dee7e7fe96200159134716e617a6e
85855 F20110217_AACFTH landera_m_Page_53.jpg
3820657cad8b121a09e8466ea7ce8a05
0ca721559f95af2a83aa11720086aff556dae744
35833 F20110217_AACFOK landera_m_Page_32.pro
fad749e4690f29f122f5dbef0ff8fb97
976b3bdc5009fec679169c11a6051f97530497de
74385 F20110217_AACFYF UFE0014823_00001.mets
e6ef85da4f9c52ab7e6af8b09cc74ef7
c3a70214eac4e970cb1445d07235649ee02c9c26
F20110217_AACFJO landera_m_Page_22.tif
c2c849b14131368ef3a7342d92c69e32
b2c8b0a7280f525ca6c13706a3b6f8b52a3f0630
28251 F20110217_AACFTI landera_m_Page_53.QC.jpg
390f79af073ffe3e8a7e84c6dc3dc134
c6d92fa8807509a8ab44c6da5ff4f4f391fbf55e
24202 F20110217_AACFOL landera_m_Page_33.pro
9dac95f458f579cf0b9b5ae70f128fbd
aa5c87f3b0b191ba9b7ef70cabf71286b38b6818
F20110217_AACFJP landera_m_Page_23.tif
7cdde45de8ffbc0b8b54e57fe4c63693
4942dcba8fe5604465ff01ae0d12b20f8aa8ed2a
101229 F20110217_AACFTJ landera_m_Page_54.jpg
fc9095f56690547b384f50d35792783b
84de91cc1a04edab4a51b8b31f2e3eea5cd48d6b
34702 F20110217_AACFOM landera_m_Page_34.pro
312f1b4cc80f2619ff31cbb26f59f012
34bb5844940aecbd401a8d129b23c0b6db6eab25
F20110217_AACFJQ landera_m_Page_24.tif
4177af1f1e1332c8d345c713e85cecc4
474d32755fc86686eee2de700a05ec1848f42a28
30748 F20110217_AACFTK landera_m_Page_55.jpg
4599be6ab99ed4233cdc254b7ab872d6
40909c0549b148594d0d7003c9a380ae597b9a99
55220 F20110217_AACFON landera_m_Page_35.pro
cb6ca21c70295a16920d2045346525cb
45cbc56217435c82fc17c65a4f4544e40c3c4014
F20110217_AACFJR landera_m_Page_25.tif
3dbea87c48dfeb72e17122d511d04c47
12eed36a8eb00296d3982a812fcb7db99235ca90
10741 F20110217_AACFTL landera_m_Page_55.QC.jpg
2dd19a784bcdc6266b981e3048e97396
9a38e6ce86008474c02a8ce754f6306da0ec1acd
33700 F20110217_AACFOO landera_m_Page_36.pro
38dc499cd790564c8dfed00487cb064f
dbf88d839da392193ffc80756ba862a7fdd6ecd4
F20110217_AACFJS landera_m_Page_26.tif
851bb885366c697a9fdc3c7d8e086190
1540664f466f004e86c1a651b2dfd8ee27a3f3ea
80468 F20110217_AACFTM landera_m_Page_56.jpg
b3fb2aed6cddbdc33ee857ee99d821f9
1ade844737a97ef319455042338b203ef78582a9
23283 F20110217_AACFOP landera_m_Page_37.pro
3cd93353957b9a46eeb31f62126f7893
e96a4211c9ba7b602c1f8b192f06d11e4dbdd066
23422 F20110217_AACFTN landera_m_Page_56.QC.jpg
5838de4236f427ea13991cfe719ebefe
0981852a0c5559d4c2870f696d02346cb74523c9
42666 F20110217_AACFOQ landera_m_Page_38.pro
d380927285fa032384a9f0364c4ddf9e
703d20e14aa18f78f72c53f0702de8531936109b
F20110217_AACFJT landera_m_Page_27.tif
8b0549dffa3bc3bf1fc82ab38f4ccab4
d6953ad9a5b13814825a6889d78bd8c9d424c937
104800 F20110217_AACFTO landera_m_Page_57.jpg
12d4d2e760d38eb37467c961e54b1727
7e2aeee2b65c0a65bcd9ec2d6a97ebf80f5dd837
14784 F20110217_AACFOR landera_m_Page_39.pro
8653547a86fead454e76eca4d6907db1
2e3a3dcf498b3339ee8011dbdc02224076062bf6
F20110217_AACFJU landera_m_Page_28.tif
595252b44cb965b87fa746cf3c89053d
fc12fc487c2195919bded8e1e00bf9c58afb8de5
29594 F20110217_AACFTP landera_m_Page_57.QC.jpg
5384ca8d4e30ac32e8b9eeb05fecfba0
bb38e880ed842989aded9350ca5a31567ebb947f
50055 F20110217_AACFOS landera_m_Page_40.pro
a2d1fe02b4d87a8b54b24561b31b4044
76ff1f580599946f6220e2941e8b83d43e04a7b2
126248 F20110217_AACFTQ landera_m_Page_58.jpg
83881d5cbe7b7dce938385ecd385b50f
832962b901f090d4c38a42ca1b05ee4889ebbf4f
52398 F20110217_AACFOT landera_m_Page_41.pro
cbdbd37efd12501a3d0c411b3dfacf80
25c162a23214600f1f82b0835784d320cf65ef21
F20110217_AACFJV landera_m_Page_29.tif
83558b96e785b5b411f1c92750adff5d
72c42b12ff29bb7c150e39414be37011f24ee3bb
36435 F20110217_AACFTR landera_m_Page_58.QC.jpg
21d3397633e9e0f1087b35958a17a894
a192362ae2b701f9efb10b08e5913d5e6d06e27e
45657 F20110217_AACFOU landera_m_Page_42.pro
d6e2b88d85b879cc26797960f9058b4a
d295124deaddfe634aae5e45da0978beeaf68105
F20110217_AACFJW landera_m_Page_31.tif
a9762266ad544da3b907f5e06aeede8d
bf017f7ad8625af0c7ca2da39954da1ddf1b329c
41301 F20110217_AACFTS landera_m_Page_59.jpg
612639919de0784cd55fed855bc28e7c
81ce267b47fab9c84fb3273618fdea9a9dfa12cc
33756 F20110217_AACFOV landera_m_Page_43.pro
60505e90ec206229cfb2d3cadf370ce8
0067ed6cda221bc787328e06fe523c675921bbe3
F20110217_AACFJX landera_m_Page_32.tif
b645de30c2092bd956818592b58204d9
6c3426810322dd5d30d3ff6acf403f02f99d18a1
11405 F20110217_AACFTT landera_m_Page_59.QC.jpg
4a031c3a1a9ab3ecdf1d7d6e9ffca3ed
f8ec3507dabe607d1401193c14088d412c509c01
45222 F20110217_AACFOW landera_m_Page_45.pro
d0ef290b3bfbe06dcfa5d012c12ddf21
28c9c192a95cc7f94fb050c8c2055a531a30f5a3
F20110217_AACFJY landera_m_Page_33.tif
53938e1edf4e19cda5ac3a2f954888a3
5521c0416612a83ea75e3f57ccdfd98b1a59f28e
30191 F20110217_AACFTU landera_m_Page_60.QC.jpg
723113a11e6fadf69e772d7fb428bc40
73b3c7ec073ecb9c1cbfc7ca506d696d2f2d0202
51717 F20110217_AACFOX landera_m_Page_46.pro
6ae28c27e88a546e4bd4f96425882fd1
54a495ece7c3689f741ec43dcb2d2a5816d9ebe2
F20110217_AACFJZ landera_m_Page_35.tif
8629d279ce0765e386b03290d69074a7
04456d57ff906386f4e1a8f20348aaeef6150109
36136 F20110217_AACFTV landera_m_Page_61.jpg
f8bccc872952c23fbfe3768095523a5e
e2151914910f5764bfd2fbe5caf3bb6c62a305c1
12158 F20110217_AACFTW landera_m_Page_61.QC.jpg
552437f3f92d94616585615e588244ae
66f95e2d0019f8c609ebd8aa280bb9a5d7b8037b
2063 F20110217_AACFMA landera_m_Page_27.txt
63fda53ff9aacf56c83de859255996c5
0560715062448b0fca7cdd93548bdd5382e01463
51982 F20110217_AACFOY landera_m_Page_47.pro
54d654717284fbbaeca7d4c0daa69ee0
92cf755e942a7ee998d4c133516d6476927fbf42
222423 F20110217_AACFTX landera_m_Page_01.jp2
3b528ca6ab06d030d72161635520d4d0
7eb632df8939d16e1a03cff2c99370efd918db40
2049 F20110217_AACFMB landera_m_Page_28.txt
fa98294aede4c299a2ddd09a812d6fef
ab4f2170d4f7fb953633b3b54277264172dfb5ae
50412 F20110217_AACFOZ landera_m_Page_48.pro
9887eed556af7b3631290404ad2a4bca
3a8a4641ee5a0f435686cc8eeb76328847a9ee09
31250 F20110217_AACFTY landera_m_Page_02.jp2
96780d6d51a36af4e22b0072dd39b447
1ec804428bd1a0db5eec0da5cae282ca8521c266
1942 F20110217_AACFMC landera_m_Page_30.txt
9b5c070dc48a63130842772658ff8304
bf189d011b4be6864746262137c12cb18ab5c9b7
888476 F20110217_AACFTZ landera_m_Page_03.jp2
d5d516bb8a1a4586cd9edbf50cd7d7d4
b5053b3a5f6ed34edffc7425ddb24346057be39e
2041 F20110217_AACFMD landera_m_Page_31.txt
e88b9485ed79fd9ff65fafa0a624b7f2
17b1cc5b3df4f7eebb9605c3e751006218227283
32466 F20110217_AACFRA landera_m_Page_21.QC.jpg
1714c6844d38a6042491cd12e43c8e4b
6f2a87b979c2d18a2e893d4485467bebe408dca6
1509 F20110217_AACFME landera_m_Page_32.txt
a079f7d22e174977114af8913c05c164
f7c672846d956e7326e40b1d896cc891f38e69ba
100604 F20110217_AACFRB landera_m_Page_22.jpg
1161a4b941175dfc2d75b5eb7a2c85cf
5d4354fb7064e572122e9367857debefc9f6a659
961 F20110217_AACFMF landera_m_Page_33.txt
5bd3bcfede62a3dce8ebd8d7a416a5e8
fe7c7a805c0db8d52d12fd452d31fda290bb7bd6
33293 F20110217_AACFRC landera_m_Page_22.QC.jpg
18dea91fb57e26a6e902dfd98efb1b5c
e2d0f7c28f6e27983a2e14c940fbe4cdb6d632d3
1590 F20110217_AACFMG landera_m_Page_34.txt
67dfd0a4dc2d0addc424483e3a834852
5b7c2a5ac736e3f369e4fafcb0a12d8a96856c05
1001232 F20110217_AACFWA landera_m_Page_60.jp2
53d3055c957b85adb4f2e84773bc2c9c
be3177a683e1ed12ec4f56840ee68c432d2291a4
97873 F20110217_AACFRD landera_m_Page_23.jpg
c7f089f3c86a229eff64f829ffbcf93a
04cbac907cbbcc725ceb42c48fd60f5bc4015d2d
2743 F20110217_AACFMH landera_m_Page_35.txt
59c2d65cae03e888f1ecca9ee5b83152
52d3780c73513938725ec94bc173a79497f7ddc4
367826 F20110217_AACFWB landera_m_Page_61.jp2
11e54b1436e2c283c6a1a10140233d18
15ca19b8cb3856691c9a41fa90ff2430d784cfa1
32252 F20110217_AACFRE landera_m_Page_23.QC.jpg
9852b7ef851491b05a91864fc4424882
83aaee04a1ccf90411be3716840be8cef9dca735
1789 F20110217_AACFMI landera_m_Page_36.txt
e0931f7a6938058c2045ef348ce05e5e
4e4a902bf8b77a79c94e47c9045f6066fac8a0e5
2170 F20110217_AACFWC landera_m_Page_01thm.jpg
6f7776722eaf6721e481c76c93c199f7
3e00bf7f1b4eb1004db5f9ede290cd818baf26f0
30068 F20110217_AACFRF landera_m_Page_24.QC.jpg
47a8a367caeaa76bf9fd58ca18cba1f8
ae4f536823f0ded8bbd115f39f3d3af9de143ec7
719 F20110217_AACFMJ landera_m_Page_39.txt
0ff44b4d6f616d8c06f20a2da4e6d339
269a96abacbb8018ae3162054b4793534c7295ec
637 F20110217_AACFWD landera_m_Page_02thm.jpg
b6a9d4fe623a9ceb059f1c7ecfebd10f
25cecdd25d016ab67817d3d98f33dc6680247d9d
83291 F20110217_AACFRG landera_m_Page_25.jpg
5b4d201328f821bfb007dc26757c430a
8c069b4b77fcf5b6a0e8b43dd1c2e999e6213447
2019 F20110217_AACFMK landera_m_Page_40.txt
69a4a35510ffd0e660a485074913af90
d184e7d2eb1e586bff6ada550e08f49132419d8e
7005 F20110217_AACFWE landera_m_Page_03thm.jpg
8b8b27656e780bbe078d7429f2a5cb67
0b118f34ca459693a70c650bee5cb9b508ed5d22
26610 F20110217_AACFRH landera_m_Page_25.QC.jpg
d1a61a4b57f94c8062ba92f3d1e966e0
f6c7197e2c09dc5f8d7e8d539956fe75b60ea917
2375 F20110217_AACFML landera_m_Page_41.txt
57a355c9453ca598ed24ed3e2e90ee04
f344ce233ba2a2bc0afe1e8f0aa5dc7b9955bc71
3844 F20110217_AACFWF landera_m_Page_04thm.jpg
d201718000702a6e797bcda6e40df6b9
8504ded46629d8de5d0b188cc72ff0b71a09b38c
115346 F20110217_AACFRI landera_m_Page_26.jpg
f228252cae6e8df497ed17f534473615
9f5cb349e088512b050e67755ee05b16d67bc08d
2072 F20110217_AACFMM landera_m_Page_42.txt
810c34375f0770eb5ce6c710eec33205
47461fc4eaa5d3373f13f1392bac5c6e5feb514b
1236 F20110217_AACFWG landera_m_Page_05thm.jpg
32061e2fdc6db5f9d8c63bdcabf79594
c4369666632f02c49b440bfe097c870ed9a89c1c
35182 F20110217_AACFRJ landera_m_Page_26.QC.jpg
f804043ee5819d85cec90a5abeb358cc
514a357f26ee43a70d7920c6d7e7c6153a9d6997
1523 F20110217_AACFMN landera_m_Page_43.txt
563f0157197a20431cd6772ef709d9ed
dddc732657198ad01a964f12b946f408e2eef298
2840 F20110217_AACFWH landera_m_Page_06thm.jpg
e9cb9d12acddabada5f1bf14239fa0e5
32eafa2ebe9985eebd18f98e2cd32c9e51056269
1920 F20110217_AACFHQ landera_m_Page_38.txt
34672b876b63c3a4d8344491a97d3414
4b891d334e1826f1a4cec43e7a6ab19fd6e4a6ca
105210 F20110217_AACFRK landera_m_Page_27.jpg
9d84e822464ffaf11cf7458f83e114b2
5693455b71c999f51bdf7e70925ff06b55cb6df2
1995 F20110217_AACFMO landera_m_Page_44.txt
84e6bf06129d27993d7b2f1eb843e819
6e29b2787f5f82e1148afbe64a59ed02b515fee6
2972 F20110217_AACFWI landera_m_Page_07thm.jpg
378878be006838d1af76765083504ee7
3b4770f3215ec156d46f2ccab3efb15092b6c6a2
34253 F20110217_AACFRL landera_m_Page_27.QC.jpg
77754295103527c58e946caf20ce3acf
2b3a4b1a65ab3e95668a1245840e9650dd156c40
1864 F20110217_AACFMP landera_m_Page_45.txt
0f268461dc6159e52e37583c2fbc4c13
4069760f1e9ebb916a578949e0efccadf28fd008
6905 F20110217_AACFWJ landera_m_Page_08thm.jpg
8d55f9b814d45a1d254e1f9dd6328084
b0a486630bd63f3f315f0a65cdd4353fc52eca0c
44883 F20110217_AACFHR landera_m_Page_44.pro
cc01c64a3f9d4e59be85366148113015
cf26580022b85e51d2fed8aff93a335c97309cc4
34412 F20110217_AACFRM landera_m_Page_28.QC.jpg
10fccf2d845843162dcfa33a04d11c9d
6a12d2158bd6bfe743c63f7aa19cb28a9f997190
2034 F20110217_AACFMQ landera_m_Page_46.txt
2eb17d62086157096e9558e20d0c63e9
cb87731e1145840c00e476dfafd9ae371de31479
2150 F20110217_AACFWK landera_m_Page_09thm.jpg
45cb161365a8515d29335a0ee479ec6b
12f0b8ecc07bfebcbf0233b097d60450ceddd4aa
2651 F20110217_AACFHS landera_m_Page_55thm.jpg
606ee08a15b9d0e3105ffc211c4638a3
ea1638b7403b3745d8c4c0645f093ca9318c81b7
101912 F20110217_AACFRN landera_m_Page_29.jpg
9f0a58df11ed0dad104d04ab511b068b
1a993663c65833c64419a4e9a0250b7abcc25778
2043 F20110217_AACFMR landera_m_Page_47.txt
81778f99539dc3502a7fcd65b13c8dcf
f46e41ce1f4f8594b4f9ad7da9cfeb7a7800ec12
7128 F20110217_AACFWL landera_m_Page_10thm.jpg
1e3ebd8593326ece9e1a17acd9871076
8e86cc68dab2a3570d1672704344d202ddd84a6c
4263 F20110217_AACFHT landera_m_Page_33thm.jpg
46782c63fb29deedd2537483b4c44342
db034f7143b2104f7e6d7e3a9c5cf90d5cac564b
33631 F20110217_AACFRO landera_m_Page_29.QC.jpg
067596043ed64d24bf4120793e433c15
35324cc5b5d2303cccd57c03366f25a0ba10b621
8223 F20110217_AACFWM landera_m_Page_11thm.jpg
c0c226ec4889dc4617469d1455046d69
ec6b38017a1cf70686d8a9e98c573bb33aee9aba
931780 F20110217_AACFHU landera_m_Page_44.jp2
8650e3baaad0fe2c5f3619bed4929478
35efb411778fc2ed77eca4acf2fa37a765bebb8c
97281 F20110217_AACFRP landera_m_Page_30.jpg
cbe586b55ab495150c45911e2f5fc5f3
256baebe0e5a48275afa9b24ea1db8881e945417
1985 F20110217_AACFMS landera_m_Page_48.txt
3dcb8fef5fb957d40acf88812ddd1afe
7e918c9be2e3edfa5726be4c4441efc9957cc262
8215 F20110217_AACFWN landera_m_Page_12thm.jpg
1bf282d0164639ba6147b002108d1364
1a597e00d1268da8c2b531ccd01970240036d9d8
632861 F20110217_AACFHV landera_m_Page_37.jp2
b0274de56bca9bc80d5fccb7f6133695
f74f4c318ba725b555cc7b69b317587a03d1c8a4
32944 F20110217_AACFRQ landera_m_Page_30.QC.jpg
00cf79757167ec2851b7911bfb417a73
b6d7503d43ca94c1970c279602aca73fba569609
2015 F20110217_AACFMT landera_m_Page_49.txt
9737e112d445a41a3d676b1eab2c39b8
842c24f5af831adf6bb4ea641a1625e881437eca
7591 F20110217_AACFWO landera_m_Page_13thm.jpg
b72444644a18d4401eac2f9d8d53d5b0
adbca1fafce9f8cfcc62c3f1c7e851c407fcd811
92241 F20110217_AACFHW landera_m_Page_60.jpg
b7e5b1e3364a39887e9c7b98939eaf71
9a35388a1e58bdc5a9cd563508dba1ccf24501b2
104462 F20110217_AACFRR landera_m_Page_31.jpg
62fbc14db8d0033103d90aca4ca19c48
4cc858a2fb3e631874e83d755eb64b4ac04e2c4c
F20110217_AACFMU landera_m_Page_50.txt
c6896651d2d7051ca431ac4dd4ddfc0d
7b4570e66765c61fda15302c1d974c52a5bcf9e8
5742 F20110217_AACFWP landera_m_Page_14thm.jpg
9e9c5fd6d3f16bbbb16bc27d70f0bea2
dbac6d3f306cb9fe0716f2d7e8f6ee592bb20d12
1088 F20110217_AACFHX landera_m_Page_37.txt
633624d397d94ea224bcaa80d2f58a0a
0dcedbf6f5f9550f2c65b85f4098547252255ef4
34172 F20110217_AACFRS landera_m_Page_31.QC.jpg
5ca278d91ab74710b8c1fc770878c8bf
2040fb6da7d65d3c17637b8ed807cd734710c1bf
F20110217_AACFMV landera_m_Page_51.txt
b9137e0fc39f89a8b5f56832d1abc155
6edfe01850933173b3b82000b7eb70fd27a1793c
7044 F20110217_AACFWQ landera_m_Page_15thm.jpg
7d30a091eb0dd95b1f59e6597f282fa7
0968992c0d49ca6054ff50451880f9aca0467e02
8484 F20110217_AACFHY landera_m_Page_50thm.jpg
d473c8e00b1e09ed708e23d55cfa9671
94c21a62ce7298f7118731761f60fe1a5c099e80
75183 F20110217_AACFRT landera_m_Page_32.jpg
4a4ac92bd44b1ec77a63593e3d3b0a9e
bfd4b89b608844deab1ca829d82034e4eb1f7cdc
7905 F20110217_AACFWR landera_m_Page_16thm.jpg
9779186c5762d378ced0bea610b8bc08
012631595bf8750ceb0c41a847e9cd42c36fea54
1984 F20110217_AACFHZ landera_m_Page_29.txt
691a14d5bb5fcab3176af618b5753c8b
72a090a41beb4346b508da791b5d838f4787fbdc
23562 F20110217_AACFRU landera_m_Page_32.QC.jpg
50bc65fda0df5af5a6171eafd67bc5cb
ed4b7cd7352b7381f9f3c5df7d88a26ac215ee2c
1800 F20110217_AACFMW landera_m_Page_52.txt
609ea4ad478f370e28ce8ca87e8e5013
2bb4d4005ef0e327dc3bb17a7f7f2b528258bc3a
8314 F20110217_AACFWS landera_m_Page_17thm.jpg
0fe182be606dfc1a8bf70d3ddfaa67e3
fec86a038e55ab96d6269a8c3767001ebe412676
52638 F20110217_AACFRV landera_m_Page_33.jpg
da38134303771a63c4fedeca6fccda3b
b6de112b73e62bdad12f0ea097bcba060f61dd56
1751 F20110217_AACFMX landera_m_Page_53.txt
9f45cdd59802196d859453bed4cc7dad
6d46b3e21a0eb229e6bc4900db67802eac5dd57c
8327 F20110217_AACFWT landera_m_Page_18thm.jpg
948ac85da471ec931a99a61e1dc111d8
e244e98a587abd6d961a5f54aa7e54d276c7de67
17362 F20110217_AACFRW landera_m_Page_33.QC.jpg
b2f762b823dfb446c5a2851a44f8bfc9
f80cc3927b44800896466265bc15a1bfa9a894e5
F20110217_AACFKA landera_m_Page_36.tif
ff5273a921439a0d4619dd4ade019f3f
a35cd4a2e2f2ae2871dcedd4aaf03b8f20fb4451
1982 F20110217_AACFMY landera_m_Page_54.txt
e98deaf210b54bd8ae1e18787e22ff0e
45189da55b1348f3e030cdc7f61f43a694e21b9b
8079 F20110217_AACFWU landera_m_Page_19thm.jpg
41727297748ff096ae39f3e5b0de5d5f
6841e1b8a0b6964ebc6a1a087d5797fafdc26405
68344 F20110217_AACFRX landera_m_Page_34.jpg
717edff1ce0df3a40a69f4955ae5a9de
8d4a12aaab918bf619237d8eaafe61b3b91529c0
F20110217_AACFKB landera_m_Page_37.tif
0ece69518a1957dbc7025a97108f7701
d77120a3c368da09201ec7024291328e78aba906
593 F20110217_AACFMZ landera_m_Page_55.txt
d0d441e56e78a6d860f3228302f51cac
998c8f0e606e5538b81a61e6c39c89f2d519fa4b
8452 F20110217_AACFWV landera_m_Page_20thm.jpg
8076bf33c66d7a54ac19db6d9b089a82
8a30a6b3dcece6e92bc125f822f5dcff17c179e1
22686 F20110217_AACFRY landera_m_Page_34.QC.jpg
58823a5c7524f6b83599ed95c3ed412f
d0add4609c3d61fad95a962ad7c93046d06fb67f
F20110217_AACFKC landera_m_Page_38.tif
1539a957e1d2bfd079b327acb8147f50
2584d43ccd959a04053cec74a49d71a14a4a10a9
8058 F20110217_AACFWW landera_m_Page_21thm.jpg
5f68013355cae2d383b27bfb4446e386
bd89b28cc5fd91cc3a652be3b4fefc13ebaafb6d
51154 F20110217_AACFPA landera_m_Page_49.pro
1935b2c824b16d2c8cd353016bffd1c2
129d2f2bb8f87afa8426da9017935a7ba35b3317
89013 F20110217_AACFRZ landera_m_Page_35.jpg
74b6a73bf6c1c50a4111dc0fdda25fc6
05b41e2672b37d764b3fc6d49982bc1bfc3327d6
F20110217_AACFKD landera_m_Page_39.tif
d0cda91f8975b7f7b99b88b44b82843d
23d4a51a84be08fc04c7c21780043a30dccffc7e
8498 F20110217_AACFWX landera_m_Page_22thm.jpg
960128898cd2c73ee6d4ac6c042566e6
fa49cfc42ea203e98c682f233c582c041f8642c9
50923 F20110217_AACFPB landera_m_Page_50.pro
f25cc6988565a5e01589abaced288676
efc9a1c1745aad2d6d9fe00a72e0412fe1ed8bf8
F20110217_AACFKE landera_m_Page_40.tif
bbb8199f3adb2a8ba6826bc5b67134c9
b00f174230da8075b69dbe06d17709e8553dd22b
7548 F20110217_AACFWY landera_m_Page_23thm.jpg
d6a4f9b073a15d6fbd7bde8ff71af413
97e527c3a9daa2d05a44c5f63893ffbd1a37fe38
50746 F20110217_AACFPC landera_m_Page_51.pro
247e41ef9a7f99f66ddb1522e0651c7a
0b1825bb08271a209dd7b95a346d017fe00892cf
F20110217_AACFKF landera_m_Page_41.tif
f6ff0c7c6e94274823c9c33adc209ca3
d91504b15449169c8c708f1f6a2d473ce0720700
7467 F20110217_AACFWZ landera_m_Page_24thm.jpg
94d5b05bbe9ac8f07e99543743fca4a2
2e9abe01694b84f95c6a3112ab2950752ceeab88
45370 F20110217_AACFPD landera_m_Page_52.pro
500290c047fefff07875c4a5f8f67f80
e5dc9aa0ed842f7e9ce79f5eaa41bd3f36e4968f
F20110217_AACFKG landera_m_Page_42.tif
472db4c2116b4145bb9ff4a87a84b462
14d169851087457819aa124508110c910a22ba67
133913 F20110217_AACFUA landera_m_Page_05.jp2
e5e0b6618ad94f9bb462277ffe6e7e5e
eb5562bce571697cf105ec4b1640de3bdd000170
42334 F20110217_AACFPE landera_m_Page_53.pro
e26760f5c42aebbeaa813694a514ea47
07928a8a3510040a274aee8083f59b981dcac0c6
F20110217_AACFKH landera_m_Page_43.tif
2d6aa9ef6d193b753e1f60e6773a7846
f318c99f1a943cf9d36f00d80cbe887094060eea
343860 F20110217_AACFUB landera_m_Page_06.jp2
e90b5306887d9f7969870bfac29bc44a
94bcf473e0f8c2de0384edd1eca2d2f7884f105c
50191 F20110217_AACFPF landera_m_Page_54.pro
59a5c1d369ffb48ba4a15123560730bb
6a304b9e99eaf6e90717db212963afa7fa061a60
F20110217_AACFKI landera_m_Page_44.tif
6232c9ad841c021fc0a6e109fb8a183b
44f725d1293dc8a38f4f1ac0c47d46e547c280f3
383203 F20110217_AACFUC landera_m_Page_07.jp2
9a4d02300df667cc7d601bef0f9fa924
348e754376d77bf58b767bdc1a8cdf6f624c01d7
40889 F20110217_AACFPG landera_m_Page_56.pro
7a7a0587fbbe73c88cc83a4d2c942730
db2745f846579ca86bb8198dcca88451d999bd00
F20110217_AACFKJ landera_m_Page_45.tif
1ef3a6162787a239d95d20d77c2f2510
6d8432e710d8e206060917dd8718d3757af6fbd6
935090 F20110217_AACFUD landera_m_Page_08.jp2
dac639e00edb246fcde83d90c9f8597b
2d2c81d987df443fc7b93749523f058df02bdab7
50998 F20110217_AACFPH landera_m_Page_57.pro
f2378d2695d701bc56e0fe1e83a5c8f3
b256ba64240bd875e4adb2f2aed62ab23b2f04ce
F20110217_AACFKK landera_m_Page_46.tif
826fd6418c40ba2e8ee03097787eede9
7213a415fb70cddd474b0dbb5a8a280d67ab21fa
964167 F20110217_AACFUE landera_m_Page_10.jp2
e5d8bd7a7b8af0a681b2937510f22646
1d91cb5954ad8243dcb1df1872966e9905e0f982
65444 F20110217_AACFPI landera_m_Page_58.pro
87d81869e622783e73fac672f462e7c0
570f1b0ff5d95609402a18933853c31295148be8
F20110217_AACFKL landera_m_Page_47.tif
3ecf99dd59d51085cbaf0e46634aaee4
d76deaa3b2163f85209edfd15c38325670db8f72
1051915 F20110217_AACFUF landera_m_Page_11.jp2
d06f80b29b4117fa3d9b323635d5e6d4
b21d1ff39ed355b0aa454df529df8ebaf29df196
17773 F20110217_AACFPJ landera_m_Page_59.pro
c9d46211b6a0111de1f5ba9da56c9888
345a754357e2fb34095912d4cd3a8f370f30fb7e
F20110217_AACFKM landera_m_Page_48.tif
ae63496eb6b884de7df24717e578b273
89f87e9d4e7a6a1a6a70818a3a757f3fd0fb426a
1051960 F20110217_AACFUG landera_m_Page_12.jp2
9b619cf93e9d8a1bc56e53da0438286e
4c25a0f3fb1cf4ae8db56d5a151673d121711c1f
44433 F20110217_AACFPK landera_m_Page_60.pro
fe2c823fdc0317ebb27a2b3ed71f2bd0
92351e6918bb89e5f4167bcf255769a9c690aed8
F20110217_AACFKN landera_m_Page_49.tif
34e5292e5a93ad928c5f4fa65f9dc4d5
cdb4a4f9e789a2af0cb451e132066d808d92436b
1019121 F20110217_AACFUH landera_m_Page_13.jp2
b213c4ac3962efc166bd8e61b59031ed
807fd055a21f7189075a9e54168e79037bf5b67a
16375 F20110217_AACFPL landera_m_Page_61.pro
5516d2bdaa196d3e3e11fe1ea86c0690
099f3d7efc56abd990ca6d3385ffb1fc68c3bbda
F20110217_AACFKO landera_m_Page_50.tif
b19970eafb4fb5277b95f7994e5b1cab
e30e72607d2783df4d166a6016953419c6354ab2
765611 F20110217_AACFUI landera_m_Page_14.jp2
6cc63a487ab72b4f9cd8c11628f36b14
34af10aae952175f90c83c7e4bd418edcf5afa7a
23626 F20110217_AACFPM landera_m_Page_01.jpg
85f8b8022f555a601c66d08ceaa1d1be
29d7f1260ba4212b780d30c39f7dba0a334d0430
F20110217_AACFKP landera_m_Page_51.tif
f40d7d95ad4876ee411e56e47ae318fe
5ba6b72f8a16ead9a873b15d706a63a98bc483a3
932530 F20110217_AACFUJ landera_m_Page_15.jp2
1e0247c757161c48094d133d0deec085
a5602457b71b5e47eb58890bf5caf1f042d3f7b0
7511 F20110217_AACFPN landera_m_Page_01.QC.jpg
b80919597f37ff504ad7941703aea8ef
85e62ad0070599ea44bacb3217f2fa20397f0807
F20110217_AACFKQ landera_m_Page_52.tif
3e20b999f5c219b952e634428010e2fc
b9f3c30b7b154ff7d69eee4fa1bf360532ba96d3
F20110217_AACFUK landera_m_Page_16.jp2
8da4d97e0c478c6c2ed99c1af9c01c81
5ef91bb5ac0b9daafc228de9b8447ce6f43d1f3a
F20110217_AACFKR landera_m_Page_53.tif
d3bf11e884d7e111f3f8f8c55d865119
9d1c6c470dba253934a94b76925e09a9ec058cbf
1051978 F20110217_AACFUL landera_m_Page_17.jp2
0b2f514614f413495629f0d53bcb032f
576a01edb8d08337f1eced20414ee2fe95d9b1b6
1733 F20110217_AACFPO landera_m_Page_02.QC.jpg
0b824c061ccb312417864963d9ad5fe3
6ce461c73affefe5c5cf866dcc68bd4ea9ccae04
F20110217_AACFKS landera_m_Page_54.tif
fd45f6548087a158cae67865980ea132
9611531ff5ccea0cc5b0c6199b7ceef6aa3c7a7c
F20110217_AACFUM landera_m_Page_18.jp2
dd3859f20ae392a8b97a08c31163336b
65dd62aa0334dd6ce713616e954a8122264e9b30
83601 F20110217_AACFPP landera_m_Page_03.jpg
e295d79bccc36a8f5ffb3daea9c5b32e
2fff148bdc45f93ec9ecb26c29f0a8538639b4c5
F20110217_AACFKT landera_m_Page_55.tif
ffa3141c2e16f8eb724267fb4007a737
0188dabaf38c9c2eeeaf7ecc975c96eb82af516f
1051954 F20110217_AACFUN landera_m_Page_19.jp2
c6ed173b91530e54087ab2b672c836a4
aca8725019e2223372f3365c33405901b546f6f7
27498 F20110217_AACFPQ landera_m_Page_03.QC.jpg
5927687d5eceafc326f8864d914aba0f
947bca8e94008cd794778e8ccdcabb54b3da5e18
1051972 F20110217_AACFUO landera_m_Page_20.jp2
c5f96e2da92b281a31f9fb673cda7a9e
e36ca83ab39a81d4fab27c83b4e767e53eaa2031
65015 F20110217_AACFPR landera_m_Page_04.jpg
d5a52c3ec4350b2376fb964413476e32
9fa3ad4fe509e81c1b27e83298171672167549ed
F20110217_AACFKU landera_m_Page_56.tif
231e57af0062244c05a04f47c468079e
4866409908908b2aacea3bf927a46ef39ae6d837
F20110217_AACFUP landera_m_Page_21.jp2
fb3b04f5e65b45b4c1b93fdb05a7191a
625bd5a2eea077cc76b9a50aac5e2953b22e51c4
15290 F20110217_AACFPS landera_m_Page_04.QC.jpg
d02a39c79126482ce1ac9baf10977ed3
1e3547e141417370e0efa6baad8ac35269e7e58f
F20110217_AACFKV landera_m_Page_57.tif
d53a2c8a2bf9adc888d083f28562897b
281ca71f5dcdd463a8c64a2d347591c1f9970185
F20110217_AACFUQ landera_m_Page_22.jp2
f492f4a05b16c4b4c6767472d27e003a
f23de03bc2466abef25d84b3650959e79741df18
16115 F20110217_AACFPT landera_m_Page_05.jpg
d0a5cdcdb757915234c23932cff55216
6ffaf6a290ed88ab718610e2b5a5d929e7223096
F20110217_AACFKW landera_m_Page_58.tif
8313baaec3af4adb1cbf75ed866967e4
0d23dcd128d6c025e560d5f678aff3d4c7c56ee3
1051969 F20110217_AACFUR landera_m_Page_23.jp2
88a42baeec2da50d0ba03bf3ae6bc3f9
e34be6703d491455283eee22e21e25e2cbb2d6b2
4518 F20110217_AACFPU landera_m_Page_05.QC.jpg
bb86390b6c59131d89e76169d7be65ce
26ae7f8f79bd9576a23d04e6ed8c29881fa1904d
F20110217_AACFKX landera_m_Page_59.tif
38785671e22f6f1a739630a57c140127
1e1146f146b5968c4ae368f34a49b11737df9ade
1004096 F20110217_AACFUS landera_m_Page_24.jp2
49b6259dd88d16f809ed6c34f63f68ed
3113dbf8be76e5f3f5af44487f1dea3df7fbb1d0
34334 F20110217_AACFPV landera_m_Page_06.jpg
1ff24b3a2047285f7f8c316662af45de
a2449e0c079c56594bc49bcbdffbde9ecf6ff32b
F20110217_AACFIA landera_m_Page_30.tif
1ba7663c0cc599fed52a8a9f6394ad7c
5d3619503155e2cc876a411f8e5463f2befc3f13
F20110217_AACFKY landera_m_Page_60.tif
edae906b5bafab3d503511351c72de58
1189435f89a5cb023720f12baa956f9745b951b5
894002 F20110217_AACFUT landera_m_Page_25.jp2
aa3d37fdaf31b4ec9aa4da7309318730
3a406184a9f61b8eef9fc084643804e0ad5b1c50
9850 F20110217_AACFPW landera_m_Page_06.QC.jpg
a8a961fee9fcd477d5d4b56b87e1344c
51ad178b75ceca36bae0a062412d5307183f138f
F20110217_AACFKZ landera_m_Page_61.tif
e9777d0b4c577e9690d884510fe4f51e
30b2891a31ed8efad566db3585941d54cacffe3c
1051984 F20110217_AACFUU landera_m_Page_26.jp2
c1451cbff56bb20b1ee859610912d88c
f515cde7a3e558f4d1b4d20beb80632823ebd90c
36414 F20110217_AACFPX landera_m_Page_07.jpg
fae7f6d0d48b49f7cdf892507a01e3fb
f7da801a2dde0774a7e51244a58a0f157a06707c
87255 F20110217_AACFIB landera_m_Page_42.jpg
342c257abdf5fb97ac4bde487add5021
6a99a2fd7620a70c4f1ff23e48dc09cea26d6922
1051926 F20110217_AACFUV landera_m_Page_27.jp2
b9ef5ca5186399cb5868cb731f9d10eb
1ea3998f86bf9cc2ba870742bdc237cc0e34e420
10728 F20110217_AACFPY landera_m_Page_07.QC.jpg
276931a720ae36f4735ad3dc50de36d5
34017904215b347f2fa9922e653e1bcc9782b74b
8311 F20110217_AACFIC landera_m_Page_48thm.jpg
d5ef7ee506185e00cd019c9c65d6b253
2e3b2730ae05120da47604559fdc1c64495a2e73
1051971 F20110217_AACFUW landera_m_Page_28.jp2
a4a5fb82b6dd47564dbf813795aa575a
77a1a2d57178b40811eca691a2080600bfed6ca7
1933 F20110217_AACFNA landera_m_Page_56.txt
7347784d7de38a74316ac4b7e3625ae1
e821af955750fb7692e346544c77ba9b7235db25
92542 F20110217_AACFID landera_m_Page_24.jpg
d2fdcbaff7993958d20bdec929650a0f
d0ff71b4faa3deef085209da51b661aead455071
F20110217_AACFUX landera_m_Page_29.jp2
4942adc12e025458ec32fcbd1176d3ae
21bd380500cc09bf2177759c5e3d2e5840990e92
2065 F20110217_AACFNB landera_m_Page_57.txt
e4edddde19c1d2c6248d99977f850f85
dec3aca2d17910f8ae8d70028c1f6e2f7bafabe0
87584 F20110217_AACFPZ landera_m_Page_08.jpg
13a754457a7ac7b4162bd5414699fc64
35b5f50685dfcfde49e074f28ec6884ec50df5a6
33064 F20110217_AACFIE landera_m_Page_49.QC.jpg
068583dcfacd59ba5cdb3f5853aafd0e
d0123cd62525ba5dc91c26519b4d8fd302a0c21a



PAGE 1

EFFECTS OF SPECTRAL SLOPE ON PERCEIVED BREATHINESS IN VOWELS By MARIO ALBERTO LANDERA A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Mario Alberto Landera

PAGE 3

ACKNOWLEDGMENTS First of all, I would like to thank Dr. David Eddins and my lab mates, Sona and Arturo, for helping me generate and organize the stimuli used in this experiment. Next, I would like to thank my committee member, Dr. Christine Sapienza, for her input in finalizing my thesis. She has also been one of my favorit e professors in my academic career because she has an ability to communicate her knowledge effectively. I would also like to thank my committee chair, Dr. Rahul Shrivastav, for guiding me throughout the research process in this ex periment. He has been a wonderful mentor to learn from and I could not have done it without him. A special thank you goes to Dr. Donna L undy. She has guided me throughout my college career in my journey towards becoming a speech-language pathologist. She is my role model and someone I aspire to become one day. If it was not for her, I would not have converted from being a Seminole to being a Gator. I also have to thank my friends Darin, Jo rge, and Javier for being there through all of my ups and downs throughout my graduate studies. They are the greatest friends I could have asked for. I would also like to thank my family for their constant love and support in every decision I have made in my academic career. They have been my backbone throughout my life and I love them all very much! Lastly, I would like to thank the National Institute for Health for providing a grant (NIH/R21 DC006690) to make this research possible. iii

PAGE 4

TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iii LIST OF TABLES.............................................................................................................vi LIST OF FIGURES..........................................................................................................vii CHAPTER 1 INTRODUCTION........................................................................................................1 2 REVIEW OF LITERATURE.......................................................................................6 Perturbation................................................................................................................... 6 Measures of Aspiration Noise......................................................................................8 First Harmonic Amplitude..........................................................................................11 Spectral Slope or Tilt..................................................................................................12 Perceptual Model for Breathy Voice Quality.............................................................13 Summary.....................................................................................................................14 Purpose.......................................................................................................................14 3 METHODS.................................................................................................................16 Listeners......................................................................................................................16 Stimuli........................................................................................................................ .16 Perceptual Ratings......................................................................................................20 Statistical Analyses.....................................................................................................21 Acoustical Analyses....................................................................................................22 4 RESULTS...................................................................................................................25 Listener Reliability.....................................................................................................25 Effects of Spectral Slope on Breathiness Ratings.......................................................26 Acoustic Analyses......................................................................................................31 Summary of Results....................................................................................................35 5 DISCUSSION.............................................................................................................36 6 CONCLUSIONS........................................................................................................44 iv

PAGE 5

APPENDIX DESCRIPTION OF PARAMETERS USED TO GENERATE TEN VOWEL STIMULI.....................................................................................................47 LIST OF REFERENCES...................................................................................................48 BIOGRAPHICAL SKETCH.............................................................................................51 v

PAGE 6

LIST OF TABLES Table page 3.1 Intra-rater reliability for the CC and VC series........................................................25 3.2 Inter-rater reliability for the CC series.....................................................................26 3.3 Inter-rater reliability for the VC series.....................................................................26 3.4 Overall listener mean rati ngs and standard deviation with increasing spectral slope.........................................................................................................................29 3.5 Relationship between H1* H2* and mean rating for each stimuli in both CC and VC series...........................................................................................................32 3.6 Total RMS power and mean ratings for ten base harmonic signal stimuli..............33 3.7 Spectral moments for ten base noise signal stimuli.................................................35 vi

PAGE 7

LIST OF FIGURES Figure page 3.1 Mean breathiness ratings for th e male speakers in the CC series............................27 3.2 Mean breathiness ratings for the female speakers in the CC series.........................27 3.3 Mean breathiness ratings for the male speakers in the VC series............................28 3.4 Mean breathiness ratings for the fe male speakers in the VC series.........................28 3.5 Relationship between listeners mean breathiness ratings and spectral slope variation for the CC series........................................................................................30 3.6 Relationship between listeners mean breathiness ratings and spectral slope variation for the VC series.......................................................................................30 3.7 Example of gender differences in the power spectrum............................................34 vii

PAGE 8

Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Arts EFFECTS OF SPECTRAL SLOPE ON PERCEIVED BREATHINESS IN VOWELS By Mario Alberto Landera August 2006 Chair: Rahul Shrivastav Major Department: Communica tion Sciences and Disorders Previous studies have indi cated that breathiness is co rrelated with measures of perturbation, aspiration noise, signal-to-noise ratio, first ha rmonic amplitude, and spectral slope. However, the role of spectral slope on perceived breathiness remains unclear. In a recent study, it was observed that varying spectr al slope resulted in minimal changes on the perceived breathiness for synthetic vowel s. However, the stimuli tested in this experiment included a relatively narrow range of spectral slope vari ation. The goal of the present experiment was to verify the role of spectral slope changes on the perception of breathiness by testing stimuli that had a wider ra nge of variation in spectral slope and a constant signal-to-noise rati o. Ten voices (5 male and 5 female) representing various levels of breathiness were synthesized using a Klatt-synthesizer. Each of these voices was manipulated to generate two continua varying in their spectral slope from -3 dB/octave to -30 dB/octave. One continuum (CC series) had a constant cutoff frequency of 500 Hz, while the other continuum (VC series) ha d a cutoff frequency between the second harmonic (H2) and the third harmonic (H3) Ten listeners judged the degree of viii

PAGE 9

breathiness using a 7-point rati ng scale. Results indicated that spectral slope had a negligible effect on the perception of breathine ss for the stimuli tested in this experiment. Furthermore, listeners rated male stimuli to be more breathy than the female stimuli in both CC and VC series. The results may be e xplained on the basis of the partial loudness model. ix

PAGE 10

CHAPTER 1 INTRODUCTION Breathiness is a term that is often used to describe a persons vocal quality. Fairbanks (1940) describes breathiness as occurring when the vocal folds fail to completely approximate during vibration, caus ing a steady stream of air that rushes audibly through the glottis and supralaryngeal tract. A breathy voice quality usually sounds soft and weak in nature, making it difficult to produce loud sounds. This can create a problem in the communication abilit ies of an individual with a breathy vocal quality, in that it draws attention to itself a nd because listeners may not be able to hear or understand what is being said to them. A breathy vocal quality can be heard in individuals with voice disorders as well as in healthy individuals. So me of the conditions that le ad to a breathy vocal quality include vocal nodules, bowing, unilateral vocal fold paralysis, psychogenic disorders, Parkinsons disease, and other neurological im pairments. Breathiness can also occur as a normal voicing characteristic. Research has sh own that females tend to have a breathier voice than males. This is due to the fact that females tend to ha ve a greater posterior glottal gap than males, allowing greater ai r to escape during p honation (Klatt & Klatt, 1990; Hanson 1997). As an individual gets older, vocal fold atrophy may occur, which results in a small glottal gap during phonation, also leading to an escape of air (Colton & Casper, 1995). Lastly, certain languages and cultures, such as Gujarati and Hmong, use a breathy vocal quality as a di stinctive feature for some of their phonemes (FischerJorgensen, 1967; Huffman, 1987). 1

PAGE 11

2 Defining and describing vocal qualities, su ch as breathiness, are generally based upon perceptual judgments. A perceptual judgment is a result of a listeners interpretation of an acoustic signal. These judgments are often first made by individuals with a vocal pathology or by the people that surround them. Perceptual judgments play an important role in how voice clinicians commonly categorize a voice condition and plan a course of treatment and/or management for their patients. For clinical purposes, perceptu al judgments are often made using a specific scale. Different kinds of scaling procedures may be used to rate an indi viduals voice quality. Each type has a specific use, with its ow n advantages and disadvantages. A clinician may want to use a categorical rating when he or she is only concerned with labeling a voice condition to a sp ecific category, such as breat hy, rough, or hoarse. A numerical rating scale involves assigning a number between 0 and n to a voice, where n represents the total number of points on the scale. The ranking on this scale represents the magnitude of the vocal quality being rate d. The two most common types of numerical rating scales used are five-poi nt and seven-point ra ting scales. If a clinician decides to use a visual analog (VA) scale, he or she is required to place a mark on an undifferentiated line, often 100 mm long, to indicate the degree to which a voice contains a given quality (Kreiman, Gerratt, Kempster, Erman, & Berke, 1993). As mentioned in Hirano (1981), the GRBAS scale is an exam ple of a standardized VA scale used for rating procedures for clinical evaluation of voice quality. There are several other types of scali ng procedures, which are often used for research on the perception of voice quality. Direct magnitude estimation (DME) involves having listeners assign a number to a voice sample to indicate the degree to which it

PAGE 12

3 contains a given quality. Ther e is generally a limitless range of possible numbers, which is designated by the experimenter. There ar e two types of DME rating scales. In an anchored design, the listener is provided with referent voice samples assigned to specific magnitudes (usually in equidist ant intervals) of the given quality. In an unanchored DME, listeners are required to make thei r ratings using their own criteria as their reference. Another method is the paired comparison task, where list eners are required to compare two stimuli and judge the degree of th eir quality on some level (Kreiman et al., 1993). In order for perceptual ratings to be mean ingful, a listener must rate a voice sample in the same manner each time it is presented. Furthermore, listeners must also be consistent with other listeners in rating a voice sample to yield meaningful results (Kreiman et al., 1993). Unfortunately, resear ch has shown that perceptual judgments vary within individuals and from one i ndividual to another (Gerratt, Kreiman, Antonanzas-Barroso, & Berke, 1993; Kreiman, Gerratt, & Precoda, 1990; Kreiman, Gerratt, Precoda, & Berke, 1992; Kreiman et al., 1993; Kreiman & Gerratt, 1996; Kreiman & Gerratt, 1998; Kreiman & Gerratt, 2000a; Kreiman & Gerratt, 2000b; Shrivastav, Sapienza & Nandur, 2005). Such inconsistencies may result from a number of factors, including, a lack of a consistent theoretical framework for measuring voice quality, poorly controlled per ceptual experiments as well as differences in stimuli, instructions, methods, and statis tics used to obtain perceptual judgments (Kreiman et al., 1993; Shrivastav et al., 2005). Internal and external standards may also influence a listeners ratings, such as momentary changes in attention, fatigue, memory of previously presented stimuli, training, past experiences wi th the stimuli and or task, and other factors

PAGE 13

4 related to chance (Shrivastav et al., 2005). These fact ors introduce considerable variability in a listeners perceptual ratings. The inconsistency in listene rs ratings of various voi ce qualities mentioned above can lead to problems in both the diagnosis and treatment of a vocal pathology. For example, a novice clinician might judge a give n voice condition as being mildly breathy. On the other hand, a trained clinician might judge the same voice condition as being moderately breathy. This discrepancy may not seem to be of any important significance at first, but when it comes time to plan a course of treatment, the novice clinician may suggest some vocal hygiene techniques to fo llow, while the trained clinician may suggest a more aggressive behavioral therapy approach, such as engaging in vocal function exercises. It is also important to consider that difficulties in measuring clinical outcome in a patient may occur due to the poor intraand inter-judge reliability documented in the studies mentioned previously. The poor inte r-judge reliability also mentioned in the studies above may also lead to difficulties in communication across clinicians in regards to a particular patient. Despite the controversy as to which met hod is best in rating and measuring voice quality, perceptual judgments remain th e most common method of describing any deviancy in an individuals voice quality. As mentioned befo re, this is how individuals first recognize any change in their voices. Du e to this fact, it is imperative that voice clinicians and research scientists devise a theoretical framework to understand how listeners perceive voice quality and one th at will yield the most reliable method for quantifying an individuals voice quality.

PAGE 14

5 One way to avoid the problems related to poor intraand inte r-judge agreement is through the use of objective measures. This method is commonly used by researchers and scientists and by some clinicians. It ma y be argued that objectiv e measures result in more accurate quantification of vocal quality as it is rule-based. Objective measures can also be more time and cost efficient and more sensitive than perceptual judgments. Also, since numbers represent a measure, they can be used to document any changes and/or progress in an individuals voice quality. However, objective measures can only be successful if they can match perceptual j udgments. Unfortunately, many of the objective measures currently being used have not been found to correlate with perceptual judgments to any significant degree (Kreiman & Gerratt, 2000a). Efforts to develop objective measures that accura tely quantify perception requi re determination of the acoustic cues for specific voice qua lities such as breathiness. Several studies have attempted to examine the acoustic correlates of breathiness. These are discussed in the next chapter. The present research takes another step in this direction. Specifically, th e goal of this research was to determine the role of spectral slope in the perception of breathiness.

PAGE 15

CHAPTER 2 REVIEW OF LITERATURE The production of breathy voice quality is ultimately determined by the physiology of the vocal mechanism. As mentioned previously, when the vocal folds fail to approximate during phonation it results in an escape of air. The sound generated by the larynx is affected by the nature of the glottal closure and voc al fold vibration patterns. This provides a source of variability in the characteristics of voices, which helps distinguish and classify voice types from one another. The effects of various glottal configurations and vocal fold closure patterns have been described by several researchers, such as Hanson (1997). These experiments s howed that the amplitude of the first harmonic (H1) is related to the open quotient of the glottal cycle whereas the spectral slope or tilt is affected by the speed of glot tal closure. An incomplete glottal closure during a cycle of vibration, results in thr ee modifications, includi ng an increase in the bandwidth of the first formant, an increase in the spectral tilt of th e glottal spectrum at high frequencies, and an emergence of turbulence noise at the glottis. However, the search for acoustic cues fo r the perception of breathiness has led researchers to look at a variety of acoustic measures. The findings of these studies are summarized below. Perturbation Perturbation refers to the short-term vari ability in the signal or cycle-to-cycle variability in the voice acoustic signal (Ost rem & Fields, 2005). It may include changes in fundamental frequency (fre quency perturbation or jitte r) or changes in intensity 6

PAGE 16

7 (intensity perturbation or shimmer). Sin ce breathy voices generally have greater aperiodicity, these measures have been hypothe sized to be related to the perception of breathiness. There are several algorithms to quantify perturba tion, and these vary in their methods for quantifying perturbation. This ma kes it difficult to compare results from experiments that have used di fferent algorithms. However, in general, experiments find a positive correlation between the perturbation in a signal and its perceived breathiness. Eskenazi, Childers, and Hicks (1990) examined six acoustic parameters, which have been shown to be good predictors in ex amining vocal quality, to determine which of these parameters were most important in pred icting five different vocal qualities, one of them being breathiness. Listeners were as ked to rate the ove rall excellence of 50 normal voices and 23 pathological voices produ cing the vowel /i/ using a 7-point rating scale in terms of various voi ce qualities. The results of this study indicated that frequency perturbation (jitter) was the mo st important predictor for a breathy voice quality. Martin, Fitch, and Wolfe (1995) analyzed two perturbation meas ures (jitter and shimmer) on eighty voice samples of the vowel /a/ representing healt hy and pathological voices. Listeners were asked to classify th e voice samples as normal, breathy, hoarse, and rough and to rate the severity of these samples on a 7-point rating scale. The results of this study indicated that le ss jitter and more shimmer were associated with the severity of breathy voices. Hillenbrand, Cleveland, and Erickson ( 1994) evaluated the effectiveness of signal periodicity in determining br eathy voice quality. Using an unr estricted direct magnitude estimation scale, listeners were asked to rate the level of breathiness of recordings of

PAGE 17

8 nonpathologic male and female speakers pr oducing normal, moderate, and very breathy vowels (/a/, /ae/, /i/, and /o/) Acoustic analysis of the ra tings on these voices revealed that signal periodicity, as measured by the cepstral peak prominence (CPP) was the most important parameter in predicting perceive d breathiness. Hillenbrand and Houde (1996) extended the same methods and examined the ability of signal peri odicity measures to predict the breathiness in disordered voices during sustained /a/ vowels and connected speech. Twenty listeners were asked to ra te the breathiness of sustained vowels and connected speech using an unbound direct magnitude estimation procedure. They found that the best predictor of br eathiness were measures related to signal periodicity (cepstral peak prominence-smoothed (CPPS), cepstral peak prominence (CPP), and Pearson r at autocorrelation peak (RPK)). Measures of Aspiration Noise Aspiration noise is referred to a turbulen t flow of air thr ough the glottis that produces an audible sound dur ing phonation (Ostrem & Fiel ds, 2005). Several studies have found aspiration noise to be a significan t predictor of breathiness. Since breathiness results from an incomplete glottal closure, these voices have a greater degree of aspiration noise. The amount of noise in th e voice is quantified using a number of methods such as the harmonic-to-noise ratio (HNR), signal-to-noise ratio (SNR), and the normalized noise energy (NNE). In general, these algorithms measure the ratio of the amplitude of a harmonic signal to the amplitude of a noise signal, and are often expressed in decibels. It is believed that voices that have more noise than harmonic energy are perceived to be breathy. Klatt and Klatt (1990) synt hesized and analyzed male and female voices to determine which acoustic parameters were mo st important in pred icting a breathy voice

PAGE 18

9 quality. Ten female and six male participants produced two sentences consisting of differing patterns of stressed and unstressed sy llables. The /a/ vowel was then extracted from these sentences for analysis. A KL SYN88 formant synthesizer was used to synthesize this vowel into natural sounding male and female voices. Listeners were then asked to determine the degree of breathiness in a pair of vowels us ing a 5-point rating scale. The results of this study indicated that aspiration noise was the most important acoustic parameter in determining breathiness. This may be due to the fact that aspiration noise occurs when the vocal folds fail to completely approximate during phonation, leading to a breathy vocal quality. Shrivastav and Pinero (2005) aimed to c onfirm the claims made by Klatt and Klatt (1990). In this study, ten listeners were asked to rate the brea thiness of vowel /a/, using a 7-point rating scale. The results of this study confirmed that aspiration noise is a significant contribu tor to perceived breathiness. Wolfe, Cornell, and Palmer (1991) inves tigated the relationship between acoustic measurements, one of which was HNR, and sp ecific voice types. In this study, the vowels /a/ and /i/ were recorded from 51 patients (20 males and 31 females) receiving voice therapy. Listeners were instructed to rate these vowels usi ng a categorical rating scale, one of which referred to breathine ss. HNR acoustic measurements were made from four different spectral regions. Spect ral Region 1 (SR1) included the first formant frequency and ranged between 0-1000 Hz. Spectral Region 2 (SR2) consisted of the second and third formants and consisted of a frequency range between 1000-3500 Hz. Spectral Region 3 (SR3) consisted of the frequency range between 3500-5000 Hz. Finally, Spectral Region 4 (SR4) consis ted of the frequenc y range between 5000-8000

PAGE 19

10 Hz. Results indicated that a breathy voice was characterized by harmonic dominance in SR1, while noise dominance was found in SR2, SR 3, and SR4. This helps illustrate the variations in HNR that occur in a breat hy voice across several frequency ranges. In another study, Martin, Fitch, and Wo lfe (1995) analyzed the HNR on eighty synthesized samples (19 males and 61 females) of the vowel /a/, representing normal and pathological voices. Listener s were asked to classify the voice samples as normal, breathy, hoarse, and rough and to rate the seve rity of these samples on a 7-point rating scale. Perceptual listening te sts indicated that a lower HNR ratio was associated with the magnitude of breathy voice quality. Similarly, Wolfe and Martin (1997) inves tigated the influence of several acoustic parameters on the prediction of severity am ong several dysphonic voice types. In this study, one of the acoustic parameters examined was SNR and one of the dysphonic voice types studied was breathiness. Fifty-one pa tients (20 males and 31 females) receiving voice therapy were asked to produce the vowel s /a/ and /i/ Listeners were asked to classify each voice type according to several dysphonic qualities and then to rate the severity of each vowel on a 7-point rating scale. Results indicated that a lower SNR produced significant correlations with a breathy voice quality. de Krom (1995) also examined the rela tionship between listeners perception of breathiness with several acoustic parameters one of which was HNR. In this study, voice fragments were recorded in seventy-ei ght speakers representing male and female voices, consisting of healthy and disordered voices. Three vowel fragments were extracted from the voice fragme nts. Listeners were then asked to rate the level of

PAGE 20

11 breathiness in the stimuli presented to them on a 10-point rating scale. The results of this study indicated that a lower HNR was the be st single predictor of breathiness. First Harmonic Amplitude The amplitude of the first harmonic is rela ted to the general shape of the glottal pulse, in particular its open quotient (Hanson, 1997). The amplitude of the first harmonic refers to the intensity, expresse d in dB, of the first harmonic of a given signal, while open quotient refers to the proporti on of a period during which the glottis is open, expressed in percentage (Klatt & Klatt, 1990). The studies mentioned below have found the first harmonic amplitude and open quotient to be si gnificant predictors of breathiness. Klatt and Klatt (1990) studied whether the first harmonic amplitude of a signal was an important acoustic parameter in predicting a breathy voice quality. The authors were able to confirm this by indicating th at the amplitude of the first harmonic was significantly correlated with th e perception of breathiness. In particular, the female voices tested in this experiment were rated as being breathier than the male voices. These female voices also demonstrated a higher amplitude of the first harmonic. Hillenbrand, Cleveland, and Erickson (1994) also evaluated the effectiveness of the first harmonic amplitude in determining a br eathy voice quality. Acous tic analysis of the ratings on these voices revealed that th e first harmonic amplitude of the voices moderately correlated with pe rceived breathiness in normal speakers simulating breathy voice quality. Hillenbrand and Houde (1996) further examined the first harmonic amplitude in patients with disordered voices and found that for the sustained vowels, the first harmonic amplitude had a moderate correl ation with breathiness. However, the first harmonic amplitude was not found to be a significant predictor of breathiness in connected speech.

PAGE 21

12 Both Klatt & Klatt (1990) and Shrivastav & Pinero (2005) observed that when open quotient is co-varied with aspiration noise, it contributes to th e perception of breathy voice quality. Since open quotient affects the H1 amplitude, this may show the role of H1 amplitude on the perception of breathiness. Spectral Slope or Tilt Spectral slope refers to how rapidly the amplitudes of successive partials (component frequencies) decrease as they ge t higher in frequency in a given spectrum (Ostrem & Fields, 2005). Although the first harmonic amplitude and open quotient may also influence the spectral slope of a signal, the effects of these changes on breathiness have been discussed previously. Some studies have suggested that spectral slope may be related to the perception of br eathiness. This is often based on the finding that a slower glottal closure, frequently seen in breathy voi ces, results in an incr ease in spectral slope (Hanson, 1997). Huffman (1987) examined measures of glottal flow in vowels produced by three Hmong male speakers. The results of this st udy indicated that a gr eater prominence of the amplitude of the fundamental frequenc y relative to the second harmonic frequency had a significant correlation with breathiness. It was also indicated that shorter closedphase duration had a significant correlation with breathiness. In another study, Childers and Ahn (1995) modeled features of the glottal volume-velocity waveform, using glottal inverse filtering. Nine adult males with one of three voice types (modal, vocal fry, and breathy) were recorded while they sustained two vowels (/a/ and /i/) and produced an allvoiced sentence. Four parameters of the Li ljencrants-Fant (LF) model were analyzed, which included the glottal pulse width, pulse skewness, abruptness of closure of the

PAGE 22

13 glottal pulse, and the spectral tilt of the glottal pulse. The results of this study indicated that a breathy voice was associated with the abruptness of glottal closure. A measure of the average ratio of the lower frequency harmonic energy to the higher frequency harmonic energy (called th e soft phonation index; SPI) and measured by the Multidimensional Voice Program (MDVP; Kay Elemetrics, Inc.) has been reported to be positively correlated to breathiness (Bhuta, Patrick, & Garnett, 2004). Other experiments, such as Klich (1982) found a positive correlation between breathiness and measures of spectral tilt obtained by co mparing energy in lowand high-frequency regions. However, this experiment did not at tempt to separate the harmonic energy from the aspiration noise prior to making such comparisons. Other studies, such as Hi llenbrand (1988), did not find any significant correlations between spectral slope and breathiness. In this study, univariate relationships between perceived dysphonia and variations in pitc h perturbation, amplitude perturbation, and additive noise in synthetically generated /a / vowels were examined. The authors stated that perceptions of breathiness were not affected by the spec tral slope of the periodic component of the signals. Perceptual Model for B reathy Voice Quality Shrivastav and Sapienza (2003) hypothesized that the perception of breathiness may be related to the partial loudness of the harmonic energy when it is masked by the aspiration noise. Partial loudness refers to th e loudness of a signal when it is heard in the presence of a masker, such as noise. Accord ing to this model, a change in breathiness may occur whenever a change in the stim ulus affects the masked loudness of the harmonic energy. Therefore, changes in either harmonic energy or aspiration noise can affect the partial l oudness of a signal.

PAGE 23

14 Summary If one was to list all of th e acoustic correlates of breathi ness proposed in the studies mentioned above, there would be a list of at least four different acoustic cues related to breathiness, some of which are specific to only breathiness and others which can be correlated with other voice qualities. When examini ng the acoustic correlates hypothesized to underlie the pe rception of breathiness, one must consider the methods used in determining their conclusions. Very few of these experiments have explicitly tested the effects of each of these paramete rs on the perception of breathiness. Rather, most studies have sought to determine correl ations between certai n acoustic parameters and breathiness; however, co rrelation does not indicate causation. Correlation may just happen due to chance or by the influence of other confounding variable s not controlled in a specific experiment. The goal of the present experiment was to confirm the findings of past research by directly manipulating specific acoustic character istics of the voice. The general approach used in this experiment was similar to that used by Klatt and Klatt (1990) as well as by Shrivastav and Pinero (2005). Both of these experiments manipulated the aspiration noise and the first harmonic amplitude in voices to determine the affect on the perceived breathiness. In contrast, the present experiment manipulated the spectral slope of the harmonic energy in voices to study it s effect on breathy voice quality. Purpose The goal of the present experiment was to verify the role of spectral slope changes on the perception of breathiness. As mentione d previously, spectral slope is affected by the abruptness of glottal cl osure (Hanson, 1997). Since voice s with incomplete glottal closure often have a slower rate of glottal closure, spectral slope may be correlated with

PAGE 24

15 breathiness. Therefore, it is hypothesized that an increase in spectral slope will result in an increase in the magnitude of perceived breathiness. This experiment was done to overcome some of the limita tions of previous experiments that have studied the effects of sp ectral slope on breathine ss. First, instead of using correlation data, the present experime nt directly modified spectral slope in synthetic voices. Second, instead of using a small number and range of spectral slope variation (such as 3 stimuli varying in a pproximately 10 dB/octave used by Klatt and Klatt, 1990), the present expe riment used a larger number and range of variation in spectral slope. Two continua varying in their spectral slope from -3 dB/octave to -30 dB/octave were created using a Klatt synt hesizer (HLSyn, Sensimetrics, Inc.) One continuum had a constant cutoff (CC) freque ncy of 500 Hz to ensu re that the first formant for all stimuli was above the cut-off frequency. However, using a fixed cut-off frequency affected male and female stimuli differently in that male stimuli had a greater number of harmonics below 500 Hz as compared to the female stimuli. The other continuum aimed to solve this problem by having a cutoff frequency (VC) between the second harmonic (H2) and the third harmonic (H3) of each stimuli to ensure that all stimuli had the same number of harmonics be low this filter cut-off frequency. A listening test was performed to evaluate the effects of these changes on perc eived breathiness. Based on the partial loudness model, it wa s hypothesized that as spectral slope increases, listeners will be able to perceive a change in breathiness, particularly in the VC series, for both male and female stimuli.

PAGE 25

CHAPTER 3 METHODS Listeners Ten young-adult females served as listeners in this experiment. The mean age of these listeners was 24.1year s and ranged from 21 to 34 years. All listeners were graduate students majoring in Speech-Language Patholog y at the University of Florida. This helped ensure that all liste ners had approximately the sa me experience and exposure in listening to and rating breathy voice quality. The listeners were native speakers of American English and had normal hearing in their right ear, as evaluated by a hearing screening at 1 kHz, 2 kHz, 4 kHz, and 8 kHz presented at 20 dB HL. All listeners were paid for their participation in the experiment. Stimuli The stimuli used in this experiment we re based upon the ten synthetic [a] vowels used by Shrivastav and Pinero (2005). These base stimuli were generated using a Klattsynthesizer (Sensimetrics Inc, 1997.). The parame ters used to generate these base stimuli are shown in Table 2.1. These ten stimuli in cluded five female voices and five male voices, and represented a wide range of breathiness. In order to systematically manipulate the sp ectral slope in each stimulus, the noise from each base stimulus had to first be rem oved, leaving only the harmonic aspect of the signal. This was necessary to ensure that manipulations of spectral slope only affected the periodic energy for each stimulus, while le aving the aspiration noise of each stimulus constant and unchanged. To achieve this, tw o versions of each base stimulus were 16

PAGE 26

17 synthesized. One version was synthesized by se tting AH (amplitude of aspiration) to 0 dB and AV (amplitude of voicing) to 60 dB. This re sulted in the synthesis of a vowel with no aspiration noise. Furthermore, OQ (open quotient ) was set to 30% and TL (tilt) was set to 15%. The second version of the same vowel was generated by setting the AH to 50 dB but setting AV to 0 dB. This resulted in a vowel with no ha rmonic energy, but one where the formants were excited using the aspira tion noise alone. This approach provided the harmonic spectrum as well as the aspiration noise spectrum for each of the ten base stimuli. Table 2.1. Parameters used to generate the 10 vowel stimuli*. ML1 ML2 ML3 ML4 ML5 FM1 FM2 FM3 FM4 FM5 F0 133.1 113.7 115.5 117.0 134.4 220.4 209.0 209.1 195.5 200.7 AV 60 60 60 60 60 60 60 60 60 60 OQ 40 55 65 75 85 40 55 65 75 85 SQ 200 200 200 200 200 200 150 350 200 200 TL 0 10 20 30 40 0 10 20 30 40 FL 10 10 10 10 10 10 10 10 10 10 AH 35 40 50 60 80 35 40 50 60 80 FNP 180 180 180 180 180 180 180 180 280 180 BNP 1000 1000 1000 1000 1000 1000 1000 40 90 30 F1 661 559 732 456 814 891 759 1050 977 957 B1 200 400 600 800 1000 200 400 600 800 1000 F2 1122 1214 1244 1187 1473 1587 1333 1470 1326 1619 B2 200 200 200 150 200 200 200 200 150 200 F3 2281 2340 2497 2463 2250 3083 2930 3000 2905 2877 B3 300 300 300 200 250 300 300 300 200 250 F4 4198 3383 3362 3405 3701 3870 4232 4000 4651 4274 B4 400 400 400 250 300 400 400 400 250 300 F5 4415 4396 4533 4194 4990 4761 4736 4990 4990 4883 B5 500 500 500 300 350 500 500 500 300 350 *ML refers to male synthetic voices and FM refers to female synthetic voices. The abbreviations on the left hand side of the table refer to the acoustic parameters in each stimulus and are standard parameters found in a Klatt-synthesizer. All abbreviations are shown in the Appendix. A series of low-pass finite impulse response 2 (FIR2) filter were generated in MATLAB 7.1 (The MathWorks Inc., 2004) to manipulate the spectral slope of the

PAGE 27

18 periodic energy for the ten base stimuli. FI R2 low-pass filters were used because they allow manipulation of the spectra l slope of a signal without affecting the other parameters of the signal. These filters were gene rated with a maximum attenuation at cutoff frequency of 1 dB, and a minimum attenuation at a high frequency of 120 dB. Each of the ten stimuli was manipulated using these filters to generate two 10-step continua varying in their spectral slope. The stimuli in each of these two conti nua varied in terms of their spectral slope in increments of 3 dB/octave, ranging from -3 dB/octave to -30 dB/octave. The first continuum included stim uli that were filtered with a fixedor constant cutoff frequency of 500 Hz. This condition is henceforth referred to as CC (constant cutoff). This condition ensured that the spectral slope for all stimuli was manipulated around at fixed cut-off frequency. The 500 Hz cut-off was selected so that the first formant for all stimuli was above the cut-off frequency. However, a fixed cut-off frequency affected male and female stimuli differently. Male stimuli, with a lower fundamental frequency, had a greater number of harmonics below 500 Hz as compared to the female stimuli which had a higher fundame ntal frequency. If the total energy in the low frequency region or the harmonic relations hips for the first few harmonics played a role in cueing breathiness, su ch differences in stimuli may af fect the final results. To further investigate this possi bility, a second continuum of stimuli was generated. This continuum was generated with a cutoff fre quency between the second harmonic (H2) and the third harmonic (H3) of each base synt hetic voiced stimuli to account for the differences between the ranges of the aver age fundamental freque ncies according to gender. This condition was labeled VC (varying cutoff). The amplitude of the first harmonic H1 has been found to be correlated w ith breathiness in past research (Huffman,

PAGE 28

19 1987). Therefore, the second stimulus continuum resulted in a series of stimuli that varied in their slope, but had the same number of harmonics below the filter cut-off frequency and had a constant H1 amplitude. A total of 200 stimuli were thus generated (10 base stimuli X 2 continua X 10 stimuli/continua). The aspiration noise for each of the ten ba se stimuli was then added to the two hundred stimuli in the CC and VC continua. Howe ver, two additional steps needed to be performed before adding the aspiration noise First, the aspiration noise for each voice was appropriately amplified to obtain a cons tant signal-to-noise ratio (SNR) of 25 dB, using MATLAB 7.1. This was essential to cr eate a proper balance between the periodic signal and the aspiration noise, so that neither of these aspect s overpowered the effects of the other. An SNR of 25 dB was chosen ba sed on pilot experiment s that showed this SNR to be ideal for the present experiment. Pilot experiment found that an average SNR of 25 dB resulted in stimuli where listeners were still able to detect differences in the voiced signal for each base stimulus. The accur acy of the algorithm used for equating the SNR was further verified by calculating the SNR for these stimuli using a commercially available software package, TF32 (Milenkovic, 1997). The SNR for all stimuli were calculated using TF32 and were found to range from 22.8 dB to 25.5 dB. However, the SNR in a single stimulus continuum was f ound to vary within a range of +/0.9 dB. Second, the aspiration noise for each voice wa s temporally shifted to align it with the filtered periodic signal The filtering for both CC and VC conditions created a 257 point shift at the beginning of the voiced signal for each stimulus. In order to properly add the noise back into the voiced signal for each stimulus, a 257 point shift was added to the beginning of the ten base noise signals. Once this wa s accomplished, the ten base

PAGE 29

20 noise signals were added back to the corresponding set of voi ced stimuli. Both of these steps were accomplished in MATLAB 7.1. Next the 257 point delay at the beginning of each new stimulus (filtered periodic signal + am plified and temporally shifted aspiration noise) was removed by deleting the zeros at the beginning of the signal using Adobe Audition 1.0 (Adobe Systems Inc., 2003). Lastly the stimuli were resampled so as to have a sampling frequency of 24,414 Hz. This wa s necessary to match the clock speed of the A/D hardware used to present the stimu li to listeners (RP2, Tucker Davis Technology Inc., 2000). Perceptual Ratings The perceptual experiment was divided in to two sessions, with listeners rating 10 randomized lists of stimuli per session. E ach session lasted approximately 45 minutes. Listeners were seated in a sound treated booth, approximately 7 ft (L) X 7 ft (B) X 6 ft (H). The stimuli were presented through an RP2 processor monoaurally through the right ear using ER-2 ear inserts (Etymotic Inc.) at an intensity of 80 dB SPL. Monoaural presentations were preferred to enable compar ison of results with other experiments that have used an auditory processing front-end to study breathiness (Shrivastav et al., 2003). Binaural integration of aud itory signals complicates certain steps in most auditory models, hence it was avoided. There is no evid ence to believe that perceptual judgments for breathiness may vary between monoaural an d binaural presentation in normal hearing listeners. Listeners were asked to rate the level of breat hiness of each voice stimulus using a seven-point rating scale, wher e a value of 1 indicated minimum breathiness and a value of 7 indicated maximum breathiness. Listeners were in structed to rate only the breathiness of each stimulus presented and to avoid making judgments based on pitch or

PAGE 30

21 loudness. No definition of breathiness wa s provided. Ratings were made using a computer monitor and a keyboard. Twenty randomized lists of voice stimuli (10 CC voice stimuli and 10 VC voice stimuli) were prepared and organized in SykofizX 2.0 software (Tucker Davis Technologies Inc., 2005). Within each list, each of the ten stimuli was presented five times in random order, for a total of 50 s timuli per list. Research has shown that averaging multiple ratings of each stimulus provides a more accurate measure of a listeners perception of voice quality (Shr ivastav, Sapienza & Nandur, 2005). Each stimulus was 489 ms in duration. Eleven m illiseconds were removed from the original signal, consisting of 500 ms, when the zeros were removed from the 257 point shift added at the beginning of each signal in MATLAB 7.1. Listeners were provided a maximum of 8 seconds to make their respons e before being presented with the next stimulus. A short break (approximately 2-3 minutes) was provided between every 3-4 lists to minimize fatigue. The five ratings obtained for each voice stimulus from each listener were averaged to obt ain a single rating. These ratings were then averaged to obtain a group mean rating for each voice stimulus. Statistical Analyses Intraand inter-judge reliability was determined using Pearsons correlation coefficient for both, CC and VC series. In tra-judge reliability was measured by determining the average correlation between each of the five ratings for each stimulus made by each listener. Inter-judge relia bility was measured by determining the correlation between each listeners mean rating for each stimulus. A linear regression analysis was perfor med in SPSS 11.0 (SPSS Inc., 2002). This was used to model the relationship betw een listeners mean breathiness ratings

PAGE 31

22 (dependent variable) and spectral slope vari ation (independent variable) for both CC and VC series. A regression function containing the y-intercept and slope for each series was created. The variance and Rsquare values for each se ries was also calculated. A two-way analysis of variance (ANOVA) wa s also performed as a confirmatory test to determine if the mean breathiness ratings for stimuli at the two ends of the spectral slope continuum (-3 dB/octave and -30 dB/octave) in each stimulus set in both CC and VC series were significantly different from each other. The ANOVA was also used to determine if any effects of gender (male vs. female stimuli) were observed. Mean breathiness ratings served as the dependent variable whereas spectral slope and gender served as the two independent variables. Any interaction between the two independent variables was also investigated. Acoustical Analyses The ten base stimuli containing only the harmonic energy (AH = 0; AV = 60) were further analyzed to determine some of their acoustic characteristics of the stimuli. This was necessary to determine differences in the perceptual judgments across stimulus series. First, the difference between the amplitudes of the first and second harmonics was calculated (H1 H2). This was done because past research has indicated that first harmonic dominance may play an important ro le in cueing breathiness (Huffman, 1987; Hanson, 1997). This would also help explai n whether the harmonic energy signals differed in the low frequency region across stim uli. The intensity of the fundamental and the second harmonics were corrected for the effects of the formant frequency using the formula described by Hanson (1997). This corr ection allows a more direct comparison of H1 H2 across stimuli varying in their fundamental and formant frequencies. The corrected H1 H2 is indicated by H1* H2* and is calculated as follows:

PAGE 32

23 )2()1(*2*1 H HHH where, H1 = Amplitude of the first harmonic, H2 = Amplitude of the second harmonic, and, = Correction factor. The value of is given by the formula: 22 1 2 110log20fFF where, F1 = Frequency of the first formant, and, f = Frequency where the harmonic is located. Another measurement to study differences across stimulus series included the calculation of total power in the high fre quency region for stimuli with no aspiration noise. This was done because the ten base stimuli differed in the overall acoustic characteristics (for example, differences in formant frequencies, formant bandwidths, harmonic density, etc.) and these affect the to tal energy in higher frequencies. To make these calculations, the stimuli were first norm alized for overall power and then filtered using a band-pass Butterworth filter. This ba nd-pass filter had cut-off frequencies of 1500 Hz and 5000 Hz, transition bands of 100 Hz and a stop attenuation of 75 dB and was generated using Adobe Audition 1.0. All base stimuli with no aspiration noise (AV = 60 dB, AH = 0 dB) were filtered and the total RMS power of the filtered signals was calculated. Finally, the characteristics of the aspira tion noise for each stimulus series were analyzed. This was done because even though the noise signals at source were held

PAGE 33

24 constant and were not manipulated in th is experiment, the various vocal tract configurations for each voice stimuli would be different and influence the formants for each voice. The noise characteristics were determined by studying the signals generated by the synthesizer with the amplitude of voicing set to zero and am plitude of aspiration noise set to 50 dB (AV = 0 dB, AH = 50 dB). These signals were first normalized for average RMS power and were then analyzed using the software TF32. To describe the nature of the noise spectrum, it was characte rized as a standard probability distribution function and its first four moments (mean, st andard deviation, skewness, and kurtosis) were calculated. Differences in these moments can be used to describe the differences in the overall shape of the noise sp ectra. The procedure used fo r this analysis was based on Forrest, Weismer, Milenkovic, and Dougall (1988).

PAGE 34

CHAPTER 4 RESULTS Listener Reliability The intra-judge reliability for each listener was determined using Pearsons correlation to examine the relationship between each of the five ratings provided by the listeners. Separate analyses were performed for the CC and VC series. For the CC series, the mean correlation for the ten listeners wa s 0.69 with a range of 0.31 to 0.97. The mean correlation in the VC series was 0.71 a nd ranged from 0.33 to 0.97. These indicate a moderately significant correlation. Table 3.1 lists the intra-judge reliability for each listener in the CC and VC series. Table 3.1. Intra-rater reliabi lity for the CC and VC series Pearsons r Listener CC VC L1 0.31 0.33 L2 0.95 0.97 L3 0.83 0.87 L4 0.95 0.96 L5 0.69 0.58 L6 0.80 0.94 L7 0.43 0.54 L8 0.81 0.83 L9 0.97 0.92 L10 0.31 0.37 Mean 0.69 0.71 The inter-judge reliability was determined by calculating the Pearsons correlations between each listeners average ratings. The me an inter-judge reliability for the CC series was 0.47 with a range of .23 to 0.91. Table 3.2 lists the inter-judge reliability between every listener for the CC series. For the VC se ries, the mean inter-judge reliability for the 25

PAGE 35

26 ten listeners was 0.55 with a range of 0.11 to 0.91. The inter-judge reliability for each listener in the VC series is presented in Table 3.3. At first glance, the interjudge reliability measures for both CC and VC series appear rather low; however, as discussed la ter, listeners did not vary much in their breathiness ratings across an increasing sp ectral slope per stimulus set. The low correlation may reflect a lack of variation in perceived breathiness across stimuli, rather than an inability of the listeners to rate the stimuli consistently. Table 3.2. Inter-rater reliability for the CC series L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L1 1 0.69 0.71 0.67 0.48 0.64 0.13 0.44 0.30 0.04 L2 1 0.87 0.91 0.62 0.85 0.05 0.66 0.63 0.39 L3 1 0.81 0.69 0.86 0.27 0.58 0.36 0.29 L4 1 0.50 0.76 0.15 0.70 0.45 0.33 L5 1 0.80 0.18 0.53 0.44 0.16 L6 1 0.21 0.68 0.46 0.35 L7 1 0.34 -0.23 0.03 L8 1 0.64 0.35 L9 1 0.35 L10 1 Table 3.3. Inter-rater reliab ility for the VC series L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L1 1 0.46 0.45 0.35 0.51 0.41 0.30 0.37 0.22 0.21 L2 1 0.81 0.84 0.81 0.91 0.60 0.79 0.35 0.68 L3 1 0.53 0.78 0.80 0.73 0.63 0.46 0.73 L4 1 0.62 0.73 0.21 0.66 0.11 0.54 L5 1 0.83 0.63 0.60 0.43 0.69 L6 1 0.61 0.67 0.34 0.65 L7 1 0.62 0.55 0.52 L8 1 0.24 0.51 L9 1 0.41 L10 1 Effects of Spectral Slope on Breathiness Ratings For the CC series, the overall mean ra ting for the male voices was 5.7 with a standard deviation of 0.14, while the mean rating for the female voices was 3.7 with a

PAGE 36

27 standard deviation of 0.5. Figure 3.1 shows the listener mean ratings for the male voices and Figure 3.2 shows the listener mean ratings for the female voices in the CC series along a continuum of increasing spectral slope. For the VC series, the mean rating for the male voices was 5.7 with a standard deviation of 0.36, while the mean rating for the female voices was 3.5 with a standard de viation of 0.30. Figure 3.3 shows the mean ratings for the male voices and Figure 3.4 shows the mean ratings for the female voices in the VC series along a continuum of increasing spectral slope. 1 2 3 4 5 6 7 -3-6-9-12-15-18-21-24-27-30 Spectral slope variation (dB/octave)Mean breathiness ratings MALE1 MALE2 MALE3 MALE4 MALE5 Figure 3.1. Mean breathiness ratings fo r the male speakers in the CC series 1 2 3 4 5 6 7 -3-6-9-12-15-18-21-24-27-30 Spectral slope variation (dB/octave)Mean breathiness ratings FEML1 FEML2 FEML3 FEML4 FEML5 Figure 3.2. Mean breathiness ratings for the female speakers in the CC series

PAGE 37

28 1 2 3 4 5 6 7 -3-6-9-12-15-18-21-24-27-30 Spectral slope variation (in dB)Mean breathiness ratings MALE1 MALE2 MALE3 MALE4 MALE5 Figure 3.3. Mean breathiness ratings for the male speakers in the VC series 1 2 3 4 5 6 7 -3-6-9-12-15-18-21-24-27-30 Spectral slope variation (in dB)Mean breathiness ratings FEML1 FEML2 FEML3 FEML4 FEML5 Figure 3.4. Mean breathiness ratings for th e female speakers in the VC series Table 3.4 lists the overall listener mean ra tings and standard deviation according to spectral slope variation for both the CC and VC series. The difference between the lowest and the highest mean ratings for the CC seri es is 0.32. The VC series demonstrates a difference of 0.48 between the lowest and the highest mean ratings. These differences

PAGE 38

29 are rather low, considering that a large cha nge (-3 dB/octave to dB/octave) was made in each stimulus continuum. Table 3.4. Overall listener mean ratings and st andard deviation with increasing spectral slope CC VC Slope (dB/octave) Mean SD Mean SD 1 4.46 1.27 4.24 1.38 2 4.50 1.24 4.38 1.29 3 4.65 1.13 4.48 1.25 4 4.65 1.13 4.58 1.23 5 4.71 1.07 4.63 1.16 6 4.76 1.11 4.67 1.17 7 4.73 1.04 4.72 1.10 8 4.77 1.08 4.67 1.15 9 4.78 1.06 4.65 1.19 10 4.73 1.06 4.70 1.17 A regression analysis was used to model the relationship between listeners mean breathiness ratings (dependent variable) and spectral slope variation (independent variable) in each series. A linear regression was performed to predict the listener mean ratings from spectral slope for both CC and VC series. For the CC series, the effects of listener mean ratings predicted by the follo wing regression functi on accounted for 73.9% of the variance in the percep tual ratings (R-square = 0.739): Breathiness Rating = 4.493 0.011 spectral slope For the VC series, the effects of listener mean ratings were predicted by the equation listed below: Breathiness Rating = 4.323 0.015 spectral slope This equation accounted for 74.4% of the va riance in the mean ratings (R-square = 0.744). Figure 3.5 and Figure 3.6 demonstrate th e relationship between listeners mean breathiness ratings and sp ectral slope variation for both CC and VC series.

PAGE 39

30 Spectral Slope (dB/octave)-3-6-9-12-15-18-21-24-27-30Mean Breathiness Ratings7.0 6.0 5.0 4.0 3.0 2.0 1.0 Rsq = 0.7386 Figure 3.5. Relationship between listeners m ean breathiness ratings and spectral slope variation for the CC series. Spectral Slope (dB/octave)-3-6-9-12-15-18-21-24-27-30Mean Breathiness Ratings7.0 6.0 5.0 4.0 3.0 2.0 1.0 Rsq = 0.7438 Figure 3.6. Relationship between listeners m ean breathiness ratings and spectral slope variation for the VC series.

PAGE 40

31 As a confirmatory test, a two-way analys is of variance (ANOVA) was performed to determine if the mean breathiness ratings for s timuli at the two ends of the continuum (-3 dB/octave and -30 dB/octave) were significan tly different from each other. Perceptual ratings of breathiness served as the depe ndent variable whereas spectral slope (-3 dB/octave or -30 dB/octave) and gender (mal e or female) served as the two independent variables. Any interaction betw een the two independent variable s was also investigated. For the CC series, no significant effect s of spectral slope were observed on breathiness ratings (F(1) = 2.719; p = 0.119). This furthe r supports the poor correlation between spectral slope breathi ness ratings in the CC series. However, a significant main effect for gender was obtained showing that the male voices were perceived to be significantly more breathy than the female voices (F(1) = 159.191; p < 0.001). No significant interaction between spectr al slope and gender was observed (F(1 ) = 0.782; p = 0.390). A significant main effect of spectral slope on breathiness ratings was demonstrated in the VC series (F(1) = 8.236; p = 0.011). This supports the slightly larger difference listeners were able to perceive between the lowest and the highest mean breathiness ratings in this series as compared to the CC series. A significant main effect of gender was obtained demonstrating that the male voices were perceived to be significantly breathier than the female voices (F(1) = 202.093; p < 0.001). No significant interaction between spectral slope and gender was observed (F(1) = 1.153; p = 0.299). Acoustic Analyses The difference between the first harmonic amplitude (H1) and the second harmonic amplitude (H2) were measured in the ten base stimuli containing only harmonic energy (AH = 0 dB; AV = 60 dB) for both the CC and VC series. This was analyzed to

PAGE 41

32 determine if the difference between the am plitude of H1 and H2 had any effects on listeners perception of breathine ss in the stimuli presented to them. H1 and H2 were not directly controlled in this experiment since these were al ways below the filter cut-off frequency. The difference in amplitude between H1 and H2 is an indication of the open quotient in a signal (Hanson, 1997) and open quotient / amplitude of H1 have been indicated as a predictor of breathiness in several studies (K latt & Klatt, 1990; Hillenbrand, Cleveland, & Houde, 1994; Hillenbrand & Houde, 1996). For the CC series, the male stimuli demons trated an H1* H2* mean of -2.8 dB, with a standard deviation of 1.21 dB. The fema le stimuli in this series demonstrated an H1* H2* mean of -4.72 dB, with a standard deviation of 2.93 dB. For the VC series, the male stimuli demonstrated an H1* H2* m ean of -1.88 dB, with a standard deviation of 1.23 dB. The female stimuli in this seri es demonstrated an H1* H2* mean of -4.86 dB, with a standard deviation of 2.87 dB. Therefore, these results indicate that on average male stimuli had a more dominant H1 amplitude than the female stimuli. Table 3.5 lists the H1* H2* with th eir corresponding mean ratings and standard deviations for each voice in the CC and VC series. Table 3.5. Relationship between H1* H2* and mean rating for each stimuli in both CC and VC series CC VC H1* H2* (in dB) Mean Rating H1* H2* (in dB) Mean Rating MALE1 -4.9 5.78 -2.9 5.47 MALE2 -2.7 5.82 -2.7 5.79 MALE3 -2.4 5.53 0.2 6.19 MALE4 -2.1 5.68 -2.1 5.76 MALE5 -1.9 5.60 -1.9 5.18 FEML1 -9.1 3.15 -9.2 3.15 FEML2 -6.1 4.04 -6.2 3.68 FEML3 -2.2 3.34 -2.6 3.33 FEML4 -2.2 4.20 -2.3 3.71 FEML5 -4 3.57 -4 3.46

PAGE 42

33 The total RMS power in the high frequency region for the ten base harmonic energy stimuli (AH = 0 dB; AV = 60 dB) was calculated to determine if any further differences were found across stimulus sets fo r both series. This analysis revealed that male voices had a mean total RMS power of .88 dB (SD = 8.18 dB) between 1500 and 5000 Hz, whereas female voices had a mean total RMS power of .14 dB (SD = 8.62 dB) in this same frequency range. There is a difference of -23.74 dB between the mean RMS power in male voices and female voi ces. The results indicate that the voicing source for the female stimuli resulted in grea ter power in the high frequency band than seen in male speakers. Table 3.6 lists the total RMS power for each of the ten base harmonic energy stimuli. Figure 3.7 demonstrat es how the spectra for the male stimuli (e.g., MALE4) and the female stimuli (e.g., FEML5) differ in the higher frequency region. This figure shows that male speakers tend to have very little harmonic energy above 2800 Hz, whereas female speakers had harmonic energy up to 5000 Hz. Table 3.6. Total RMS power and mean rati ngs for ten base harmonic signal stimuli Stimulus F0 Total RMS Power (dB) Mean Rating (CC) Mean Rating (VC) MALE1 132 -71.41 5.78 5.47 MALE2 114 -69.99 5.82 5.79 MALE3 116 -63.26 5.53 6.19 MALE4 117 -68.42 5.68 5.76 MALE5 135 -51.32 5.60 5.18 FEML1 220 -35.01 3.15 3.15 FEML2 209 -54.58 4.04 3.68 FEML3 209 -38.16 3.34 3.33 FEML4 196 -44.49 4.20 3.71 FEML5 200 -33.47 3.57 3.46

PAGE 43

34 0 20 40 60 80 100 120 0 25005000750010000 Frequency (Hz)Amplitude (dB) FEML5 MALE4 Figure 3.7. Example of gender diffe rences in the power spectrum The spectral characteristics of the noise were further analyzed in the ten base noise stimuli (AH = 50 dB; AV = 0 dB). Each noi se spectrum was treated as a probability distribution function and its fi rst four moments were calcul ated (mean, SD, skewness and kurtosis). These analyses were completed using TF32 (Milenkovic, 1997). In the male stimuli, the noise spectra had a mean of 1647 Hz, a standard deviation of 1220 Hz, skewness of 4.01, and a kurtosis of 30.06. The female stimuli demonstrated a mean frequency of 1826 Hz, a standard deviati on of 920 Hz, skewness of 2.92, and a kurtosis of 17.43. These results indicate that the male stimuli used in this experiment had a lower average noise frequency than for female stimuli. The aspiration noise in the male stimuli was also observed to be more skewed to the right than the female stimuli. Finally, the male stimuli were observed to have larger kurtosis than the female stimuli. Table 3.7 lists the spectral moments for each of the ten base noise stimuli.

PAGE 44

35 Table 3.7. Spectral moments for ten base noise signal stimuli Mean (Hz) SD (Hz) Skew Kurtosis MALE1 1006 1142 5.191 36.03 MALE2 1415 1458 3.517 18.294 MALE3 1876 1178 2.958 19.472 MALE4 2055 1595 2.552 11.611 MALE5 1882 729 5.848 64.898 FEML1 2033 997 1.791 9.347 FEML2 1538 886 3.572 23.256 FEML3 1763 824 3.471 24.844 FEML4 1672 976 3.293 16 FEML5 2126 927 2.457 13.678 Summary of Results Listeners demonstrated a moderately si gnificant intra-judge reliability in both series. However, these listeners demonstrat ed a weaker inter-j udge correlation in both CC and VC series. This may be due to the fact that an increase in spectral slope had little effect on listeners perception of breathiness. The difference between the lowest and the highest mean breathiness ratings across spectra l slope were relatively small, even though the VC series demonstrated a slightly gr eater and statistically significant difference between the two end-points of the continuum. A regression analysis supported this finding and showed a weak relationship betw een the spectral slope and breathiness ratings. Male stimuli were rated significantly higher in terms of br eathiness than the female stimuli for both series. Acoustic analyses of the stimuli showed that the male and female stimuli also differed in their H1* H 2*, the average power of the harmonics in the high frequency region and in the spectral char acteristics of their aspiration noise. These differences may be responsible for the gender effect found in this experiment.

PAGE 45

CHAPTER 5 DISCUSSION The goal of the present study was to determ ine the effects of changes in spectral slope on the perception of breathiness. This was done because the role of spectral slope on perceived breathiness remains unclear, with some studies indicating that spectral slope plays an important role in the perception of breathiness (Huffman, 1987; Klatt & Klatt, 1990; Childers & Ahn, 1995), while other studies su ch as Hillenbrand (1 988), stated that spectral slope was not associated with breathi ness. The results of this present study are discussed below. Reliability measurements were taken to determine the consistency of listeners within themselves and with one another in making perceptual ratings. Pearsons correlation revealed that the intra-judge reliab ility varied among liste ners in both the CC and VC series. The CC series demonstrated only a moderate level of intra-judge reliability (0.69). Three listeners had intrajudge reliability under 0.50. The VC series also demonstrated a moderate level of intr a-judge reliability ( 0.71). Two listeners had intra-judge reliability under 0.50. The fact that listeners were not able to perceive much of a difference in levels of breathiness in the stimuli presented to them may be a reason why they demonstrated moderately high levels of reliability. In order to obtain a high correlation between two variables, there must be sufficient variab ility in the data. If there is no variation, then the two variables will not demonstrate high levels of correlation. The inter-judge reliability also varied among listeners in both the CC and VC series. Both, CC and VC series demonstrated an overall moderate level of inter-judge 36

PAGE 46

37 reliability (Pearsons correlation of 0.47 a nd 0.55, respectively). Although these measures appear rather low, this may again reflect the small variance in the perceptual data. Therefore, the low inter-judge reliability likely results from the nature of the stimuli rather than differences across listeners. This was further confirmed by the findings discussed below. Perceptual ratings indicated that there is li ttle change in perceived breathiness when spectral slope is manipulated in both CC and VC conditions. The difference between the lowest and the highest mean breathiness ratings across spectral slope position in the CC and VC series demonstrated a differen ce of 0.32 and 0.48, respect ively. Although these differences were relatively small; the mean rating for the stimuli located at -3 dB/octave and at -30 dB/octave in the VC series were found to be statistically significant, according to a two-way analysis of variance (ANOVA). A linear regression analysis examined the relationship between spectral slope variation and listeners mean breathiness ratings in both series. For both CC and VC series, th e regression function accounted for a large amount of variance in the per ceptual data (R-squares of 0.739 for the CC series and 0.744 for the VC series). However, the slope of these regression functions were very low (0.011 and -0.015 for the CC and VC series, respec tively) suggesting that variations in spectral slope had only a sma ll effect on perceived breathiness in these stimuli. Although these results agree with some pa st research (for example, Hillenbrand, 1988), these contradict the findi ngs of some other studies th at have found measures of spectral slope to correlate with breathiness (Huffman, 1987; Klatt & Klatt, 1990; Childers & Ahn, 1995; Bhuta, Patrick, & Garnett, 2004). These differences may be attributed to certain methodological differences. The current study systematically manipulated spectral

PAGE 47

38 slope in a controlled manner. Unlike previous studies that used natural stimuli, factors such as SNR, open quotient, and first harmonic amplitude were controlled in this current study to minimize their influence on the result s. These factors have been shown to be predictors of breathiness in several studies (Huffman, 1987; Klatt & Klatt, 1990; Hillenbrand, Cleveland, & Erickson, 1994; Childers & Ahn, 1995; de Krom, 1995; Martin, Fitch, & Wolfe, 1995; Hillenbrand & Houde, 1996; B huta, Patrick, & Garnett, 2004; Shrivastav & Pinero, 2005) and these may have co-varied with changes in spectral slope. The SNR for all stimuli was held cons tant at 25 dB and the open quotient was set to 30% for every stimulus. Additionally, the SN R value of 25 dB may also partly explain why spectral slope variation did not affect breathiness in the present experiment. This is further discussed below. The findings of the present experiment ma y also be explained using the partial loudness model described by Shrivastav and Sapienza (2003). Sin ce partial loudness is related to the level of the harmonic energy rela tive to that of the aspiration noise, changes in either of these parameters can affect partial loudness. The stimuli used in this experiment varied in their spectral slope, but had a constant SNR, obtained by modifying the overall level of the harmonic energy while ke eping a constant aspiration noise level. An increase in the spectral slope without any changes to the level of aspiration noise would result in a decrease in partial loudness of th e harmonic signal. The partial loudness is also dependent on the spectral shap e of the signal and the masker. Therefore, once the aspiration noise completely masks th e harmonic signal at specific frequencies, a further change in spectral slope would have li ttle affect on partial loudness. The results of this study follow this pattern in that on average, listeners are able to detect differences in

PAGE 48

39 breathiness in a stimulus among the first two in stances of an increasing spectral slope in the CC series and among the first three instances of an increasing spec tral slope in the VC series. Presumably, an increase in spectral sl ope after these levels provides no additional masking. Thus, there is no further change in partial loudness, or in perceived breathiness. The fact that the spectral slope variation re sulted in a slightly greater increase in breathiness for the VC series may be related to the lower filter cutoff frequency in these series (particularly for the male stimuli). A lower filter cut-off frequency may affect partial loudness to a greater degree because the filtering would affect the level of the harmonic signal to a greater degree. This model would further predict that cha nges in spectral slope may have failed to affect the breathiness for these stimuli becau se the SNR of 25 dB may have already masked the harmonic energy significantly. A fu rther increase in spectral slope may not have resulted in any signifi cant change in partial loudness of the harmonic energy. This model would further predict that if the SNR were increased, a change in spectral slope would result in a greater change in breathi ness. This is because a higher SNR would result in a greater difference between the leve ls of the harmonics and the aspiration noise. A change in spectral slope for these stimuli would lead to a greater change in masking, and hence partial loudness and breathiness. However, this prediction needs to be empirically tested. A significant gender effect was also observed for the mean ratings of breathiness. As shown in Figure 3.1 and 3.2, the five male synthetic voices were rated to be more breathy (ratings between 5.2 and 5.8) than female voices (ratings between 2.7 and 4.4). Figures 3.3 and 3.4 demonstrate similar diffe rences for the VC series. A two-way

PAGE 49

40 analysis of variance (ANOVA) confirmed the ge nder differences as being significant. It is interesting that the synthetic male voices were perceived to be breathier than the synthetic female voices, since female voices have been reported to be breathier voice quality than male voices (Colton & Casper, 1995). Closer examination of the acoustic prope rties of the harmonic signals in these stimuli demonstrated several differences betwee n the male and female stimuli. First, male stimuli had a more dominant H1 amplitude th an the female stimuli. Second, calculation of total RMS power in specific frequency ba nds revealed that the female stimuli had greater harmonic energy between 1500 Hz and 5000 Hz as compared to the male stimuli. Upon examining the range of the last harmonic in the male and female stimuli, it was noted that the last harmonic in the male stimuli occurred between 1000 Hz and 1500 Hz, while the last harmonic in the female stimuli occurred between 1700 Hz and 2400 Hz. This goes along with the fact that males have la rger vocal tracts than females, resulting in lower resonant frequencies and lower formant peaks, which in turn affect the harmonic and noise signals of a stimulus. Third, the aspiration noise spectra for the male and female stimuli differed in several ways. The male stimuli demonstrated a lower mean frequency than the female stimuli. The male stimuli also demonstrated a greater skewness to the right and had a greater level of kurtosis than the female stimuli. Together, these differences in the harmonic and aspiration nois e spectra leads to a greater influence of noise in the male stimuli, as compared to the female stimuli. The kurtosis of one male stimuli (MALE5) was almost three times as large as the next highest stimuli. This voice stimulus may have this large amount of kurtosis due to its noise stimuli occurring at a low level. If this stimulus is removed, the overall

PAGE 50

41 difference between male and female stimuli is not very significant. These acoustic differences in the harmonic energy and aspi ration noise between the male and female stimuli directly affect the partial loudness patterns for the voices and can explain the gender differences observed in the perceptual ratings. The results of this experiment must be inte rpreted in light of the fact that: (1) the cutoff frequency was set to 500 Hz or between H2 and H3 of a stimulus; (2) the open quotient was set to 30%; and (3) the SNR was set to 25 dB. If th e three variables of cutoff frequency, open quotient, and SNR are va ried from the parameters used in this study, the results may differ. For example, tw o cutoff frequencies used in this current study yielded slightly different results in that the VC series demonstrat ed a slightly larger range of perceptual ratings compared to th e CC series. On the other hand, raising the open quotient to a higher percentage would in crease the amount of time the vocal folds are open relative to the total duration of the period, thus increasing the H1 amplitude. Lastly, decreasing the SNR would lead to a s timuli containing more noise than signal, leading the noise aspect to dominate the ha rmonic energy. The effects of each of these three factors needs to be empirically studied to obtain a complete understanding of how spectral slope may affect breathiness. A second limitation deals with the fact that the noise signal wa s kept constant for all stimuli. This creates a problem, as wa s discussed in terms of the partial loudness model. As spectral slope is increased, the same amount of noise could result in greater masking of the harmonic energy. However, if the harmonic levels are too low, an increase in the SNR will have no further affect on masking the harmonic energy. The steeper spectral slopes in this current study may have been pe rceived as being breathier if

PAGE 51

42 the SNR was maintained at a higher level. Futu re studies should test this possibility, as it will help shed light on the appropriateness of partial loudness in predicting breathiness. Another limitation deals with the use of synthetic stimuli. The synthetic stimuli used in this experiment only had energy up to 5000 Hz. However, natural voices may have energy (especially the aspiration noise ) extending above this range. This loss of high frequency energy in the synthetic stimuli may lead to somewhat different results as compared to natural voices. This may further affect the perceptual ra tings of breathiness. Future experiments may need to consider th e role of frequencies above 5 kHz in the perception of breathiness. The fact that only the vowel [a] was used in this study may also be considered a further limitation of this study. Other vowel s are produced with different vocal tract configurations, which may lead to different outcomes. Connected sp eech has been shown to produce some differing results when compared to vowels (Hillenbrand et al., 1996). These considerations could be addr essed in future studies. Future studies should compare breathy voi ces found in healthy individuals with breathy voices resulting from various voice disorders. Th e results of this current study differ from those of previous studies that have found spectral slope to be a significant predictor of breathiness (Huffman, 1987; Klatt & Klatt, 1990; Childers & Ahn, 1995). One reason for these differences may be the choice of stimuli in these experiments. In these studies, breathy voices found in healthy individuals were used to analyze various measures of spectral slope, while this curre nt study used voice stimuli consisting of a variety of voice disorders. Both normal a nd disordered voices, consisting of various levels of breathiness, should be examined in a future study under the same methodology.

PAGE 52

43 It may be that breathy voices observed in healthy individuals has better SNR than found in disordered voices. Examining this issue will help determine if the two groups of voices are distinctly different or whether they constitute different regions on the same continuum. Future research should also verify the role of the other acoustic correlates mentioned in previous studies. As mentioned previously, there are at least four different acoustic cues related to breat hiness. Some of these para meters are specific to only breathiness, while others have been shown to be significant predictors of other voice qualities. Many of these studi es looked for correlations be tween an acoustic parameter and the perception of breathine ss without explicitly testing the effects of these parameters on the perception of breathiness. These fu ture studies should try to incorporate a common theoretical framework that controls for every possible confounding variable, which should lead to more accurate acoustic predictors of breathiness. Once we are better able to know all of the predictors of breathiness, and other vocal qualities for that matter, clinicians will be better able to objectiv ely assess voice qualities in individuals who present with a vocal pathology. Clinicians can then use these measures as supplements to their subjective ratings of vocal qualitie s to gain a better picture of a patients voice condition. By obtaining objective meas ures, intra-rater and inter-rater reliability measures will also im prove, as objective measures would help yield more consistent measures in measuring the cl inical outcome in a patient over time and also would add more consistency in communication across clinicians.

PAGE 53

CHAPTER 6 CONCLUSIONS The effects of spectral slope manipulations for voice stimuli were analyzed to determine listeners perception of breathiness. Two continua varying in spectral slope were created. The stimuli in each continuum were filtered using high-pass filters with slopes ranging from -3 dB/octave to -30 dB/oct ave in increments of 3 dB/octave. The first continuum (CC series) contained stimuli which were low pass filtered at a constant cutoff frequency of 500 Hz to ensure that the first formant of each stimulus would not be filtered. The second continuum (VC series) co ntained stimuli which were filtered at a cutoff frequency between H2 and H3 of each stimulus to ensure that each stimulus set would have the same number of harmonics below the filter cutoff frequency. Furthermore, the open quotient of each stimulus was set to 30% and the SNR was set at 25 dB. Listeners perceptual rati ngs demonstrated that as spectral slope was increased in each set of stimuli there was little change in perceived br eathiness for both CC and VC series. This was confirmed statistically by performing a regression analysis, which indicated a very low slope value between list eners ratings from -3 dB/octave to -30 dB/octave for both series. A two-way ANOVA was also performed and indicated that the mean breathiness ratings for the VC se ries demonstrated a small but significant increase in the mean breathiness ratings for stimuli with the -30 dB/octave filter when compared to the -3 dB/octave condition. No significant increase in breathiness was observed for the CC series. 44

PAGE 54

45 A significant gender effect fo r perceptual ratings of br eathiness was also observed. In both CC and VC series, the male stimuli were rated to be more breathy than the female stimuli. This finding was confirmed st atistically through a two-way ANOVA. The acoustic properties of the harmonic signals in these stimuli revealed several differences between the male and female stimuli with th e male stimuli having greater H1 amplitude, less harmonic energy in the higher frequency, and differences in the aspiration noise spectra. Together, these differences may acc ount for the differences observed in the perceptual ratings between the male and female stimuli. The effects of spectral slope variation as well as the gender differences obtained in the present study may be explained on the basi s of changes in the partial loudness of the harmonic energy when it is masked by the as piration noise. The small effect of spectral slope variation may have resulted because of a relatively small SNR (25 dB). Based on the partial loudness model, it is predicted th at spectral slope variations would have a greater effect on breathiness for a higher SN R. However, this needs to be empirically verified. In conclusion, this study indicates that sp ectral slopes role on the perception of breathiness may be secondary to that of the aspiration noise. Unlik e previous research studies that found spectral slope to be important (Huffma n, 1987; Klatt & Klatt, 1990; Childers & Ahn, 1995), the presen t experiment found that spectral slope had a very small effect on the perception of breathiness. The di fferences in these findings may relate to differences in the other parameters for the stimuli (i.e., SNR, open quotient, first harmonic amplitude, etc.) used in different experiments (Huffman, 1987; Eskenazi, Childers, & Hicks, 1990; Klatt & Klatt, 1990; Hillenbrand, Clevel and, & Erickson, 1994;

PAGE 55

46 Childers & Ahn, 1995; Martin, Fitch, & Wo lfe, 1995; Hillenbr and & Houde, 1996; Bhuta, Patrick, & Garnett, 2004; Shrivastav & Pinero, 2005). Future research should investigate the effect of other such paramete rs in a systematic and controlled manner to better understand their role on breathiness. This will result in the development of appropriate models for voice quality perception as well as tools that will allow clinicians to objectively assess individuals presenting wi th various levels of breathy vocal quality.

PAGE 56

APPENDIX DESCRIPTION OF PARAMETERS USED TO GENERATE TEN VOWEL STIMULI Parameter MIN VAL MAX Description F0 0 1000 5000 Fundamental frequency, in tenths of an Hz AV 0 60 80 Amplitude of voicing, in dB OQ 10 50 99 Open quotient (voicing open-time/period), in % SQ 100 200 500 Speed quotient (rise/fall time of open period, LF model only), in % TL 0 0 41 Extra tilt of voicing spectrum, dB down at 3 kHz FL 0 0 100 Flutter (random fluct in f 0), in % of maximum AH 0 0 80 Amplitude of aspiration, in dB FNP 180 280 500 Frequency of the nasal pole, in Hz BNP 40 90 1000 Bandwidth of the nasal pole, in Hz F1 180 500 1300 Frequency of the first formant, in Hz B1 30 60 1000 Bandwidth of the first formant, in Hz F2 550 1500 3000 Frequency of the second formant, in Hz B2 40 90 1000 Bandwidth of the second formant, in Hz F3 1200 2500 4800 Frequency of the third formant, in Hz B3 60 150 1000 Bandwidth of the third formant, in Hz F4 2400 3250 4990 Frequency of the fourth formant, in Hz B4 100 200 1000 Bandwidth of the fourth formant, in Hz F5 3000 3700 4990 Frequency of the fifth formant, in Hz B5 100 200 1500 Bandwidth of the first formant, in Hz MIN represents the minimum value of the pa rameter. VAL represents the default value which is applied if the user makes no cha nges. MAX represents the maximum value of the parameter **Table adapted from Klatt and Klatt (1990) 47

PAGE 57

LIST OF REFERENCES Bhuta, T., Patrick, L., & Garnett, J. D. (2004) Perceptual evaluation of voice quality and its correlation with acoustic measurements. Journal of Voice, 18 (3), 299-304. Childers, D. G., & Ahn, C. (1995). Modeling the glottal volume-velocity waveform for three voice types. Journal of the Acoustical Society of America, 97 (1), 505-519. Colton, R., & Casper, J. K. (1995). Understanding voice problems: A physiological perspective for diagnosis and treatment. Baltimore: Williams and Wilkins. de Krom, G. (1995). Some spectral correlate s of pathological breathy and rough voice quality for different types of vowel fragments. Journal of Speech and Hearing Research, 38 794-811. Eskenazi, L., Childers, D. G., & Hicks, D. M. (1990). Acoustic correlates of vocal quality. Journal of Speech and Hearing Research, 33 298-306. Fairbanks, G. (1940). Voice and articulation drillbook New York: Harper and Brothers. Fischer-Jorgensen, E. (1967). Phonetic an alysis of breathy (murmured) vowels in Gujarati. Indian Linguistics, 28, 71-139. Forrest, K., Weismer, G., Milenkov ic, P., & Dougall, R. N. (1988). Statistical analysis of word-initial voiceless obstruents: Preliminary data.. Journal of the Acoustical Society of America, 84(1), 115-123. Gerratt, B. R., Kreiman, J., Antonanzas-Barro so, N., & Berke, G. S. (1993). Comparing internal and external standards in voice quality judgments. Journal of Speech and Hearing Research, 36 14-20. Hanson, H. (1997). Glottal characteristics of female speakers: Acoustic correlates. Journal of the Acoustical Society of America, 101 (1), 466-481. Hillenbrand, J. (1988). Perception of aperiodi cities in synthetically generated voices. Journal of the Acoustical Society of America, 83 (6), 2361-2371. Hillenbrand, J., Cleveland, R. A., & Erickson, R. L. (1994). Acoustic correlates of breathy vocal quality. Journal of Speech and Hearing Research, 37 769-778. 48

PAGE 58

49 Hillenbrand, J., & Houde, R. A. (1996). Ac oustic correlates of breathy vocal quality: Dysphonic voices and continuous speech. Journal of Speech and Hearing Research, 39 311-321. Hirano, M. (1981). Clinical examination of voice. New York: Springer-Verlag. Huffman, M. (1987). Measures of phonation type in Hmong. Journal of the Acoustical Society of America, 81(2), 495-504. Klatt, D., & Klatt, L. (1990). Analysis synthesis, and perception of voice quality variations among female and male talkers. Journal of the Acoustical Society of America, 87(2), 820-857. Klich, R. J. (1982). Relationships of vowel characteristics to listener ratings of breathiness. Journal of Speech and Hearing Research, 25 574-580. Kreiman, J., & Gerratt, B. R. (1996). The pe rceptual structure of pathological voice quality. Journal of the Acoustical Society of America, 100 (3), 1787-1797. Kreiman, J., & Gerratt, B. R. (1998). Validit y of rating scale measures of voice quality. Journal of the Acoustical Society of America, 104 (3), 1598-1608. Kreiman, J., & Gerratt, B. R. (2000a). Measur ing voice quality. In R. D. Kent, & M. J. Ball (Eds.), Voice quality measurement (pp. 73-101). San Diego, CA: Singular. Kreiman, J., & Gerratt, B. R. (2000b). Sources of listener disagreem ent in voice quality assessment. Journal of the Acoustical Society of America, 108 (4), 1867-1876. Kreiman, J., Gerratt, B. R., Kempster, G.B., Erman, A., & Berke, G.S. (1993). Perceptual evaluation of voice quality: Review, tutorial, and a framework for future research. Journal of Speech and Hearing Research, 36 21-40. Kreiman, J., Gerratt, B. R., Precoda, K. (1990) Listener experience and perception of voice quality. Journal of Speech and Hearing Research, 33, 103-115. Kreiman, J., Gerratt, B. R., Precoda, K., & Berke, G. S. (1992). Individual differences in voice quality perception. Journal of Speech and Hearing Research, 35 512-520. Martin, D., Fitch, J., & Wolfe, V. (1995). Pathologic voice type and the acoustic prediction of severity. Journal of Speech and Hearing Research, 38 765-771. Ostrem, J., & Fields, J. (2005). Tutorials: Voice production Retrieved November 3, 2005, from The National Center for Voice and Speech Web site: http://www.ncvs.org/ncvs/tutorials/ voiceprod/tutorial/index.html. Shrivastav, R., & Pinero, M. (2005). Effects of aspiration noise and spectral slope on perceived breathiness in vowels. Journal of the Acoustical Society of America, 117(4), 2622-2623.

PAGE 59

50 Shrivastav, R., & Sapienza, C. M. (2003). Ob jective measures of breathy voice quality obtained using an auditory model. Journal of Acoustical Society of America, 114(4), 2217-2224. Shrivastav, R., Sapienza, C. M., & Nandur V. (2005). Application of psychometric theory to the measurement of voi ce quality using rating scales. Journal of Speech, Language, and Hearing Research, 48 1-13. Wolfe, V., Cornell, R., & Palmer, C. (1991) Acoustic correlates of pathologic voice types. Journal of Speech and Hearing Research, 34 509-516. Wolfe, V., & Martin, D. (1997). Acoustic corr elates of dysphonia: Type and severity. Journal of Communication Disorders, 30 403-416.

PAGE 60

BIOGRAPHICAL SKETCH Mario Landera is a graduating masters student in the University of Florida Department of Communication Sciences and Diso rders. During his masters program, he completed a masters thesis examining th e effects of spectral slope on perceived breathiness under the mentorship of Rahul Shrivastav, Ph.D., which was accepted as a poster presentation at the 151st Acoustical Society of America (ASA) Meeting. Mr. Landera received his B.S. in communication sciences and disorders from the Florida State University in May 2004. In his senior year, he completed a senior honors thesis examining social isolation in adolescents who stutter under the mentorsh ip of Lisa Scott, Ph.D., which was accepted as a poster presentation at the 2004 annual American SpeechLanguage Hearing Association (ASHA) Conve ntion. He was also recognized as the outstanding senior in speech-l anguage pathology during his senior year. Over his four years of undergraduate studies, he was honored with membership into Phi Kappa Phi honor society, Phi Sigma Theta honor societ y, Lambda Pi Eta honor society, and the National Society of Collegiate Scholars. He has also been on the Deans List for his GPA throughout his college career. Before beginning his graduate studies at th e University of Florida, Mr. Landera was accepted as a Board of Education fellow in th e summer of 2004, where he was instructed on the research process and writi ng. During his first year at th e University of Florida as a full-time graduate student, he worked as a gr aduate assistant at the Office of Graduate Minority Programs, assisting in various r ecruitment and retention tasks targeting 51

PAGE 61

52 underrepresented minority graduate students. In his second year as a graduate student at the University of Florida, he worked as a graduate research assistant in the voice perception lab in the Department of Comm unication Sciences and Disorders, under the supervision of Rahul Shrivastav, Ph.D. His duties have included a review of literature on voice quality, design of an experiment, genera ting appropriate stimuli, recruiting test participants, and data collection and analysis. In July 2006, Mr. Landera will begin his clinical fellowship year at the Miami Veterans Affairs Medical Center in Miami, Florida.


Permanent Link: http://ufdc.ufl.edu/UFE0014823/00001

Material Information

Title: Effects of Spectral Slope on Perceived Breathiness in Vowels
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0014823:00001

Permanent Link: http://ufdc.ufl.edu/UFE0014823/00001

Material Information

Title: Effects of Spectral Slope on Perceived Breathiness in Vowels
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0014823:00001


This item has the following downloads:


Full Text












EFFECTS OF SPECTRAL SLOPE ON PERCEIVED BREATHINESS IN VOWELS


By

MARIO ALBERTO LANDERA













A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ARTS

UNIVERSITY OF FLORIDA


2006

































Copyright 2006

by

Mario Alberto Landera















ACKNOWLEDGMENTS

First of all, I would like to thank Dr. David Eddins and my lab mates, Sona and

Arturo, for helping me generate and organize the stimuli used in this experiment.

Next, I would like to thank my committee member, Dr. Christine Sapienza, for her

input in finalizing my thesis. She has also been one of my favorite professors in my

academic career because she has an ability to communicate her knowledge effectively.

I would also like to thank my committee chair, Dr. Rahul Shrivastav, for guiding

me throughout the research process in this experiment. He has been a wonderful mentor

to learn from and I could not have done it without him.

A special thank you goes to Dr. Donna Lundy. She has guided me throughout my

college career in my journey towards becoming a speech-language pathologist. She is

my role model and someone I aspire to become one day. If it was not for her, I would not

have converted from being a Seminole to being a Gator.

I also have to thank my friends Darin, Jorge, and Javier for being there through all

of my ups and downs throughout my graduate studies. They are the greatest friends I

could have asked for.

I would also like to thank my family for their constant love and support in every

decision I have made in my academic career. They have been my backbone throughout

my life and I love them all very much!

Lastly, I would like to thank the National Institute for Health for providing a grant

(NIH/R21 DC006690) to make this research possible.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S .................................................................. ......... ................ iii

LIST OF TABLES ............................... .... ......................... vi

L IST O F F IG U R E S .... ...... ................................................ .. .. ..... .............. vii

CHAPTER

1 IN TRODU CTION ................................................. ...... .................

2 REVIEW OF LITERATURE ......................................................... .............. 6

Perturbation.................. ................................................. ..... ........ ..... 6
M measures of A spiration N oise ................................................................... ......... ..8
First Harmonic Amplitude ........................................................... ... ............11
Spectral Slope or Tilt ................................................................. .. .. .............. 12
Perceptual Model for Breathy Voice Quality ..........................................................13
Sum m ary ..................................... .................. ................. ........... 14
P u rp o se ............................................................................ 14

3 M ETHOD S ..................................... .................. .............. ........... 16

L isten ers ...................................... ......................................................16
S tim u li ............................................................................... 1 6
Perceptual R ratings ...................... ...................... ... ......... .... ....... 20
Statistical A analyses ................................................. .. ........ .... ... 21
A cou stical A n aly ses........... ...... ............................................................ ........ .. ....... .. 22

4 R E S U L T S .............................................................................2 5

L listener R liability ...................................................... ...... .... ................ ......25
Effects of Spectral Slope on Breathiness Ratings...........................................26
A acoustic A analyses .......................................... ............. .... ... ....31
Sum m ary of R esults......... .............................................................. .. .......... ... 3 5

5 D ISCU SSIO N ...................................................................... .......... 36

6 CON CLU SION S .................................. .. .......... .. .............44










APPENDIX DESCRIPTION OF PARAMETERS USED TO GENERATE TEN
V O W E L S T IM U L I ............................................................................ ....................4 7

L IST O F R E F E R E N C E S .......................................................................... ....................48

BIO GRAPH ICAL SK ETCH ....................................................................51




















































v
















LIST OF TABLES

Table p

3.1 Intra-rater reliability for the CC and VC series............................................ 25

3.2 Inter-rater reliability for the C C series ........................................ .....................26

3.3 Inter-rater reliability for the VC series............................................ .................. 26

3.4 Overall listener mean ratings and standard deviation with increasing spectral
slo p e ............................................................................. 2 9

3.5 Relationship between HI* H2* and mean rating for each stimuli in both CC
an d V C series s ...................................................................... 3 2

3.6 Total RMS power and mean ratings for ten base harmonic signal stimuli ..............33

3.7 Spectral moments for ten base noise signal stimuli .............................................35
















LIST OF FIGURES


Figurege

3.1 Mean breathiness ratings for the male speakers in the CC series ..........................27

3.2 Mean breathiness ratings for the female speakers in the CC series .......................27

3.3 Mean breathiness ratings for the male speakers in the VC series ..........................28

3.4 Mean breathiness ratings for the female speakers in the VC series.........................28

3.5 Relationship between listeners' mean breathiness ratings and spectral slope
variation for the C C series............................................... ............................. 30

3.6 Relationship between listeners' mean breathiness ratings and spectral slope
variation for the V C series. ........................................................... .....................30

3.7 Example of gender differences in the power spectrum ................. ................34















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Arts

EFFECTS OF SPECTRAL SLOPE ON PERCEIVED BREATHINESS IN VOWELS

By

Mario Alberto Landera

August 2006

Chair: Rahul Shrivastav
Major Department: Communication Sciences and Disorders

Previous studies have indicated that breathiness is correlated with measures of

perturbation, aspiration noise, signal-to-noise ratio, first harmonic amplitude, and spectral

slope. However, the role of spectral slope on perceived breathiness remains unclear. In a

recent study, it was observed that varying spectral slope resulted in minimal changes on

the perceived breathiness for synthetic vowels. However, the stimuli tested in this

experiment included a relatively narrow range of spectral slope variation. The goal of the

present experiment was to verify the role of spectral slope changes on the perception of

breathiness by testing stimuli that had a wider range of variation in spectral slope and a

constant signal-to-noise ratio. Ten voices (5 male and 5 female) representing various

levels of breathiness were synthesized using a Klatt-synthesizer. Each of these voices was

manipulated to generate two continue varying in their spectral slope from -3 dB/octave to

-30 dB/octave. One continuum (CC series) had a constant cutoff frequency of 500 Hz,

while the other continuum (VC series) had a cutoff frequency between the second

harmonic (H2) and the third harmonic (H3). Ten listeners judged the degree of









breathiness using a 7-point rating scale. Results indicated that spectral slope had a

negligible effect on the perception of breathiness for the stimuli tested in this experiment.

Furthermore, listeners rated male stimuli to be more breathy than the female stimuli in

both CC and VC series. The results may be explained on the basis of the partial loudness

model.














CHAPTER 1
INTRODUCTION

Breathiness is a term that is often used to describe a person's vocal quality.

Fairbanks (1940) describes breathiness as occurring when the vocal folds fail to

completely approximate during vibration, causing a steady stream of air that rushes

audibly through the glottis and supralaryngeal tract. A breathy voice quality usually

sounds soft and weak in nature, making it difficult to produce loud sounds. This can

create a problem in the communication abilities of an individual with a breathy vocal

quality, in that it draws attention to itself and because listeners may not be able to hear or

understand what is being said to them.

A breathy vocal quality can be heard in individuals with voice disorders as well as

in healthy individuals. Some of the conditions that lead to a breathy vocal quality

include vocal nodules, bowing, unilateral vocal fold paralysis, psychogenic disorders,

Parkinson's disease, and other neurological impairments. Breathiness can also occur as a

normal voicing characteristic. Research has shown that females tend to have a breathier

voice than males. This is due to the fact that females tend to have a greater posterior

glottal gap than males, allowing greater air to escape during phonation (Klatt & Klatt,

1990; Hanson 1997). As an individual gets older, vocal fold atrophy may occur, which

results in a small glottal gap during phonation, also leading to an escape of air (Colton &

Casper, 1995). Lastly, certain languages and cultures, such as Gujarati and Hmong, use a

breathy vocal quality as a distinctive feature for some of their phonemes (Fischer-

Jorgensen, 1967; Huffman, 1987).









Defining and describing vocal qualities, such as breathiness, are generally based

upon perceptual judgments. A perceptual judgment is a result of a listener's

interpretation of an acoustic signal. These judgments are often first made by individuals

with a vocal pathology or by the people that surround them. Perceptual judgments play

an important role in how voice clinicians commonly categorize a voice condition and

plan a course of treatment and/or management for their patients.

For clinical purposes, perceptual judgments are often made using a specific scale.

Different kinds of scaling procedures may be used to rate an individual's voice quality.

Each type has a specific use, with its own advantages and disadvantages. A clinician

may want to use a categorical rating when he or she is only concerned with labeling a

voice condition to a specific category, such as breathy, rough, or hoarse. A numerical

rating scale involves assigning a number between 0 and n to a voice, where n represents

the total number of points on the scale. The ranking on this scale represents the

magnitude of the vocal quality being rated. The two most common types of numerical

rating scales used are five-point and seven-point rating scales. If a clinician decides to

use a visual analog (VA) scale, he or she is required to place a mark on an

undifferentiated line, often 100 mm long, to indicate the degree to which a voice contains

a given quality (Kreiman, Gerratt, Kempster, Erman, & Berke, 1993). As mentioned in

Hirano (1981), the GRBAS scale is an example of a standardized VA scale used for

rating procedures for clinical evaluation of voice quality.

There are several other types of scaling procedures, which are often used for

research on the perception of voice quality. Direct magnitude estimation (DIME) involves

having listeners assign a number to a voice sample to indicate the degree to which it









contains a given quality. There is generally a limitless range of possible numbers, which

is designated by the experimenter. There are two types of DIME rating scales. In an

anchored design, the listener is provided with referent voice samples assigned to specific

magnitudes (usually in equidistant intervals) of the given quality. In an unanchored

DME, listeners are required to make their ratings using their own criteria as their

reference. Another method is the paired comparison task, where listeners are required to

compare two stimuli and judge the degree of their quality on some level (Kreiman et al.,

1993).

In order for perceptual ratings to be meaningful, a listener must rate a voice sample

in the same manner each time it is presented. Furthermore, listeners must also be

consistent with other listeners in rating a voice sample to yield meaningful results

(Kreiman et al., 1993). Unfortunately, research has shown that perceptual judgments

vary within individuals and from one individual to another (Gerratt, Kreiman,

Antonanzas-Barroso, & Berke, 1993; Kreiman, Gerratt, & Precoda, 1990; Kreiman,

Gerratt, Precoda, & Berke, 1992; Kreiman et al., 1993; Kreiman & Gerratt, 1996;

Kreiman & Gerratt, 1998; Kreiman & Gerratt, 2000a; Kreiman & Gerratt, 2000b;

Shrivastav, Sapienza & Nandur, 2005). Such inconsistencies may result from a number

of factors, including, a lack of a consistent theoretical framework for measuring voice

quality, poorly controlled perceptual experiments as well as differences in stimuli,

instructions, methods, and statistics used to obtain perceptual judgments (Kreiman et al.,

1993; Shrivastav et al., 2005). Internal and external standards may also influence a

listener's ratings, such as momentary changes in attention, fatigue, memory of previously

presented stimuli, training, past experiences with the stimuli and or task, and other factors









related to chance (Shrivastav et al., 2005). These factors introduce considerable

variability in a listener's perceptual ratings.

The inconsistency in listeners' ratings of various voice qualities mentioned above

can lead to problems in both the diagnosis and treatment of a vocal pathology. For

example, a novice clinician might judge a given voice condition as being mildly breathy.

On the other hand, a trained clinician might judge the same voice condition as being

moderately breathy. This discrepancy may not seem to be of any important significance

at first, but when it comes time to plan a course of treatment, the novice clinician may

suggest some vocal hygiene techniques to follow, while the trained clinician may suggest

a more aggressive behavioral therapy approach, such as engaging in vocal function

exercises. It is also important to consider that difficulties in measuring clinical outcome

in a patient may occur due to the poor intra- and inter-judge reliability documented in the

studies mentioned previously. The poor inter-judge reliability also mentioned in the

studies above may also lead to difficulties in communication across clinicians in regards

to a particular patient.

Despite the controversy as to which method is best in rating and measuring voice

quality, perceptual judgments remain the most common method of describing any

deviancy in an individual's voice quality. As mentioned before, this is how individuals

first recognize any change in their voices. Due to this fact, it is imperative that voice

clinicians and research scientists devise a theoretical framework to understand how

listeners perceive voice quality and one that will yield the most reliable method for

quantifying an individual's voice quality.









One way to avoid the problems related to poor intra- and inter-judge agreement is

through the use of objective measures. This method is commonly used by researchers

and scientists and by some clinicians. It may be argued that objective measures result in

more accurate quantification of vocal quality as it is rule-based. Objective measures can

also be more time and cost efficient and more sensitive than perceptual judgments. Also,

since numbers represent a measure, they can be used to document any changes and/or

progress in an individual's voice quality. However, objective measures can only be

successful if they can match perceptual judgments. Unfortunately, many of the objective

measures currently being used have not been found to correlate with perceptual

judgments to any significant degree (Kreiman & Gerratt, 2000a). Efforts to develop

objective measures that accurately quantify perception require determination of the

acoustic cues for specific voice qualities such as breathiness.

Several studies have attempted to examine the acoustic correlates of breathiness.

These are discussed in the next chapter. The present research takes another step in this

direction. Specifically, the goal of this research was to determine the role of spectral

slope in the perception of breathiness.














CHAPTER 2
REVIEW OF LITERATURE

The production of breathy voice quality is ultimately determined by the physiology

of the vocal mechanism. As mentioned previously, when the vocal folds fail to

approximate during phonation it results in an escape of air. The sound generated by the

larynx is affected by the nature of the glottal closure and vocal fold vibration patterns.

This provides a source of variability in the characteristics of voices, which helps

distinguish and classify voice types from one another. The effects of various glottal

configurations and vocal fold closure patterns have been described by several researchers,

such as Hanson (1997). These experiments showed that the amplitude of the first

harmonic (H1) is related to the open quotient of the glottal cycle whereas the spectral

slope or tilt is affected by the speed of glottal closure. An incomplete glottal closure

during a cycle of vibration, results in three modifications, including an increase in the

bandwidth of the first formant, an increase in the spectral tilt of the glottal spectrum at

high frequencies, and an emergence of turbulence noise at the glottis.

However, the search for acoustic cues for the perception of breathiness has led

researchers to look at a variety of acoustic measures. The findings of these studies are

summarized below.

Perturbation

Perturbation refers to the short-term variability in the signal or cycle-to-cycle

variability in the voice acoustic signal (Ostrem & Fields, 2005). It may include changes

in fundamental frequency (frequency perturbation orjitter) or changes in intensity









(intensity perturbation or shimmer). Since breathy voices generally have greater

aperiodicity, these measures have been hypothesized to be related to the perception of

breathiness. There are several algorithms to quantify perturbation, and these vary in their

methods for quantifying perturbation. This makes it difficult to compare results from

experiments that have used different algorithms. However, in general, experiments find a

positive correlation between the perturbation in a signal and its perceived breathiness.

Eskenazi, Childers, and Hicks (1990) examined six acoustic parameters, which

have been shown to be good predictors in examining vocal quality, to determine which of

these parameters were most important in predicting five different vocal qualities, one of

them being breathiness. Listeners were asked to rate the "overall excellence" of 50

normal voices and 23 pathological voices producing the vowel /i/ using a 7-point rating

scale in terms of various voice qualities. The results of this study indicated that

frequency perturbation (jitter) was the most important predictor for a breathy voice

quality.

Martin, Fitch, and Wolfe (1995) analyzed two perturbation measures (jitter and

shimmer) on eighty voice samples of the vowel /a/ representing healthy and pathological

voices. Listeners were asked to classify the voice samples as normal, breathy, hoarse,

and rough and to rate the severity of these samples on a 7-point rating scale. The results

of this study indicated that less jitter and more shimmer were associated with the severity

of breathy voices.

Hillenbrand, Cleveland, and Erickson (1994) evaluated the effectiveness of signal

periodicity in determining breathy voice quality. Using an unrestricted direct magnitude

estimation scale, listeners were asked to rate the level of breathiness of recordings of









nonpathologic male and female speakers producing normal, moderate, and very breathy

vowels (/a/, /ae/, /i/, and /o/). Acoustic analysis of the ratings on these voices revealed

that signal periodicity, as measured by the cepstral peak prominence (CPP) was the most

important parameter in predicting perceived breathiness. Hillenbrand and Houde (1996)

extended the same methods and examined the ability of signal periodicity measures to

predict the breathiness in disordered voices during sustained /a/ vowels and connected

speech. Twenty listeners were asked to rate the breathiness of sustained vowels and

connected speech using an unbound direct magnitude estimation procedure. They found

that the best predictor of breathiness were measures related to signal periodicity (cepstral

peak prominence-smoothed (CPPS), cepstral peak prominence (CPP), and Pearson r at

autocorrelation peak (RPK)).

Measures of Aspiration Noise

Aspiration noise is referred to a turbulent flow of air through the glottis that

produces an audible sound during phonation (Ostrem & Fields, 2005). Several studies

have found aspiration noise to be a significant predictor of breathiness. Since breathiness

results from an incomplete glottal closure, these voices have a greater degree of

aspiration noise. The amount of noise in the voice is quantified using a number of

methods such as the harmonic-to-noise ratio (HNR), signal-to-noise ratio (SNR), and the

normalized noise energy (NNE). In general, these algorithms measure the ratio of the

amplitude of a harmonic signal to the amplitude of a noise signal, and are often expressed

in decibels. It is believed that voices that have more noise than harmonic energy are

perceived to be breathy.

Klatt and Klatt (1990) synthesized and analyzed male and female voices to

determine which acoustic parameters were most important in predicting a breathy voice









quality. Ten female and six male participants produced two sentences consisting of

differing patterns of stressed and unstressed syllables. The /a/ vowel was then extracted

from these sentences for analysis. A KLSYN88 formant synthesizer was used to

synthesize this vowel into natural sounding male and female voices. Listeners were then

asked to determine the degree of breathiness in a pair of vowels using a 5-point rating

scale. The results of this study indicated that aspiration noise was the most important

acoustic parameter in determining breathiness. This may be due to the fact that aspiration

noise occurs when the vocal folds fail to completely approximate during phonation,

leading to a breathy vocal quality.

Shrivastav and Pinero (2005) aimed to confirm the claims made by Klatt and Klatt

(1990). In this study, ten listeners were asked to rate the breathiness of vowel /a/, using a

7-point rating scale. The results of this study confirmed that aspiration noise is a

significant contributor to perceived breathiness.

Wolfe, Cornell, and Palmer (1991) investigated the relationship between acoustic

measurements, one of which was HNR, and specific voice types. In this study, the

vowels /a/ and /i/ were recorded from 51 patients (20 males and 31 females) receiving

voice therapy. Listeners were instructed to rate these vowels using a categorical rating

scale, one of which referred to breathiness. HNR acoustic measurements were made

from four different spectral regions. Spectral Region 1 (SR1) included the first formant

frequency and ranged between 0-1000 Hz. Spectral Region 2 (SR2) consisted of the

second and third formants and consisted of a frequency range between 1000-3500 Hz.

Spectral Region 3 (SR3) consisted of the frequency range between 3500-5000 Hz.

Finally, Spectral Region 4 (SR4) consisted of the frequency range between 5000-8000









Hz. Results indicated that a breathy voice was characterized by harmonic dominance in

SR1, while noise dominance was found in SR2, SR3, and SR4. This helps illustrate the

variations in HNR that occur in a breathy voice across several frequency ranges.

In another study, Martin, Fitch, and Wolfe (1995) analyzed the HNR on eighty

synthesized samples (19 males and 61 females) of the vowel /a/, representing normal and

pathological voices. Listeners were asked to classify the voice samples as normal,

breathy, hoarse, and rough and to rate the severity of these samples on a 7-point rating

scale. Perceptual listening tests indicated that a lower HNR ratio was associated with the

magnitude of breathy voice quality.

Similarly, Wolfe and Martin (1997) investigated the influence of several acoustic

parameters on the prediction of severity among several dysphonic voice types. In this

study, one of the acoustic parameters examined was SNR and one of the dysphonic voice

types studied was breathiness. Fifty-one patients (20 males and 31 females) receiving

voice therapy were asked to produce the vowels /a/ and /i/ Listeners were asked to

classify each voice type according to several dysphonic qualities and then to rate the

severity of each vowel on a 7-point rating scale. Results indicated that a lower SNR

produced significant correlations with a breathy voice quality.

de Krom (1995) also examined the relationship between listeners' perception of

breathiness with several acoustic parameters, one of which was HNR. In this study,

voice fragments were recorded in seventy-eight speakers representing male and female

voices, consisting of healthy and disordered voices. Three vowel fragments were

extracted from the voice fragments. Listeners were then asked to rate the level of









breathiness in the stimuli presented to them on a 10-point rating scale. The results of this

study indicated that a lower HNR was the best single predictor of breathiness.

First Harmonic Amplitude

The amplitude of the first harmonic is related to the general shape of the glottal

pulse, in particular its open quotient (Hanson, 1997). The amplitude of the first harmonic

refers to the intensity, expressed in dB, of the first harmonic of a given signal, while open

quotient refers to the proportion of a period during which the glottis is open, expressed in

percentage (Klatt & Klatt, 1990). The studies mentioned below have found the first

harmonic amplitude and open quotient to be significant predictors of breathiness.

Klatt and Klatt (1990) studied whether the first harmonic amplitude of a signal

was an important acoustic parameter in predicting a breathy voice quality. The authors

were able to confirm this by indicating that the amplitude of the first harmonic was

significantly correlated with the perception of breathiness. In particular, the female

voices tested in this experiment were rated as being breathier than the male voices. These

female voices also demonstrated a higher amplitude of the first harmonic.

Hillenbrand, Cleveland, and Erickson (1994) also evaluated the effectiveness of the

first harmonic amplitude in determining a breathy voice quality. Acoustic analysis of the

ratings on these voices revealed that the first harmonic amplitude of the voices

moderately correlated with perceived breathiness in normal speakers simulating breathy

voice quality. Hillenbrand and Houde (1996) further examined the first harmonic

amplitude in patients with disordered voices and found that for the sustained vowels, the

first harmonic amplitude had a moderate correlation with breathiness. However, the first

harmonic amplitude was not found to be a significant predictor of breathiness in

connected speech.









Both Klatt & Klatt (1990) and Shrivastav & Pinero (2005) observed that when open

quotient is co-varied with aspiration noise, it contributes to the perception of breathy

voice quality. Since open quotient affects the H1 amplitude, this may show the role of

H1 amplitude on the perception of breathiness.

Spectral Slope or Tilt

Spectral slope refers to how rapidly the amplitudes of successive partial

(component frequencies) decrease as they get higher in frequency in a given spectrum

(Ostrem & Fields, 2005). Although the first harmonic amplitude and open quotient may

also influence the spectral slope of a signal, the effects of these changes on breathiness

have been discussed previously. Some studies have suggested that spectral slope may be

related to the perception of breathiness. This is often based on the finding that a slower

glottal closure, frequently seen in breathy voices, results in an increase in spectral slope

(Hanson, 1997).

Huffman (1987) examined measures of glottal flow in vowels produced by three

Hmong male speakers. The results of this study indicated that a greater prominence of

the amplitude of the fundamental frequency relative to the second harmonic frequency

had a significant correlation with breathiness. It was also indicated that shorter closed-

phase duration had a significant correlation with breathiness. In another study, Childers

and Ahn (1995) modeled features of the glottal volume-velocity waveform, using glottal

inverse filtering. Nine adult males with one of three voice types (modal, vocal fry, and

breathy) were recorded while they sustained two vowels (/a/ and /i/) and produced an all-

voiced sentence. Four parameters of the Liljencrants-Fant (LF) model were analyzed,

which included the glottal pulse width, pulse skewness, abruptness of closure of the









glottal pulse, and the spectral tilt of the glottal pulse. The results of this study indicated

that a breathy voice was associated with the abruptness of glottal closure.

A measure of the average ratio of the lower frequency harmonic energy to the

higher frequency harmonic energy (called the soft phonation index; SPI) and measured

by the Multidimensional Voice Program (MDVP; Kay Elemetrics, Inc.) has been

reported to be positively correlated to breathiness (Bhuta, Patrick, & Garnett, 2004).

Other experiments, such as Klich (1982) found a positive correlation between breathiness

and measures of spectral tilt obtained by comparing energy in low- and high-frequency

regions. However, this experiment did not attempt to separate the harmonic energy from

the aspiration noise prior to making such comparisons.

Other studies, such as Hillenbrand (1988), did not find any significant correlations

between spectral slope and breathiness. In this study, univariate relationships between

perceived dysphonia and variations in pitch perturbation, amplitude perturbation, and

additive noise in synthetically generated /a/ vowels were examined. The authors stated

that perceptions of breathiness were not affected by the spectral slope of the periodic

component of the signals.

Perceptual Model for Breathy Voice Quality

Shrivastav and Sapienza (2003) hypothesized that the perception of breathiness

may be related to the partial loudness of the harmonic energy when it is masked by the

aspiration noise. Partial loudness refers to the loudness of a signal when it is heard in the

presence of a masker, such as noise. According to this model, a change in breathiness

may occur whenever a change in the stimulus affects the masked loudness of the

harmonic energy. Therefore, changes in either harmonic energy or aspiration noise can

affect the partial loudness of a signal.









Summary

If one was to list all of the acoustic correlates of breathiness proposed in the studies

mentioned above, there would be a list of at least four different acoustic cues related to

breathiness, some of which are specific to only breathiness and others which can be

correlated with other voice qualities. When examining the acoustic correlates

hypothesized to underlie the perception of breathiness, one must consider the methods

used in determining their conclusions. Very few of these experiments have explicitly

tested the effects of each of these parameters on the perception of breathiness. Rather,

most studies have sought to determine correlations between certain acoustic parameters

and breathiness; however, correlation does not indicate causation. Correlation may just

happen due to chance or by the influence of other confounding variables not controlled in

a specific experiment.

The goal of the present experiment was to confirm the findings of past research by

directly manipulating specific acoustic characteristics of the voice. The general approach

used in this experiment was similar to that used by Klatt and Klatt (1990) as well as by

Shrivastav and Pinero (2005). Both of these experiments manipulated the aspiration noise

and the first harmonic amplitude in voices to determine the affect on the perceived

breathiness. In contrast, the present experiment manipulated the spectral slope of the

harmonic energy in voices to study its effect on breathy voice quality.

Purpose

The goal of the present experiment was to verify the role of spectral slope changes

on the perception of breathiness. As mentioned previously, spectral slope is affected by

the abruptness of glottal closure (Hanson, 1997). Since voices with incomplete glottal

closure often have a slower rate of glottal closure, spectral slope may be correlated with









breathiness. Therefore, it is hypothesized that an increase in spectral slope will result in

an increase in the magnitude of perceived breathiness.

This experiment was done to overcome some of the limitations of previous

experiments that have studied the effects of spectral slope on breathiness. First, instead of

using correlation data, the present experiment directly modified spectral slope in

synthetic voices. Second, instead of using a small number and range of spectral slope

variation (such as 3 stimuli varying in approximately 10 dB/octave used by Klatt and

Klatt, 1990), the present experiment used a larger number and range of variation in

spectral slope. Two continue varying in their spectral slope from -3 dB/octave to -30

dB/octave were created using a Klatt synthesizer (HLSyn, Sensimetrics, Inc.) One

continuum had a constant cutoff (CC) frequency of 500 Hz to ensure that the first

formant for all stimuli was above the cut-off frequency. However, using a fixed cut-off

frequency affected male and female stimuli differently in that male stimuli had a greater

number of harmonics below 500 Hz as compared to the female stimuli. The other

continuum aimed to solve this problem by having a cutoff frequency (VC) between the

second harmonic (H2) and the third harmonic (H3) of each stimuli to ensure that all

stimuli had the same number of harmonics below this filter cut-off frequency. A listening

test was performed to evaluate the effects of these changes on perceived breathiness.

Based on the partial loudness model, it was hypothesized that as spectral slope

increases, listeners will be able to perceive a change in breathiness, particularly in the VC

series, for both male and female stimuli.














CHAPTER 3
METHODS

Listeners

Ten young-adult females served as listeners in this experiment. The mean age of

these listeners was 24.lyears and ranged from 21 to 34 years. All listeners were graduate

students majoring in Speech-Language Pathology at the University of Florida. This

helped ensure that all listeners had approximately the same experience and exposure in

listening to and rating breathy voice quality. The listeners were native speakers of

American English and had normal hearing in their right ear, as evaluated by a hearing

screening at 1 kHz, 2 kHz, 4 kHz, and 8 kHz presented at 20 dB HL. All listeners were

paid for their participation in the experiment.

Stimuli

The stimuli used in this experiment were based upon the ten synthetic [a] vowels

used by Shrivastav and Pinero (2005). These base stimuli were generated using a Klatt-

synthesizer (Sensimetrics Inc, 1997.). The parameters used to generate these base stimuli

are shown in Table 2.1. These ten stimuli included five female voices and five male

voices, and represented a wide range of breathiness.

In order to systematically manipulate the spectral slope in each stimulus, the noise

from each base stimulus had to first be removed, leaving only the harmonic aspect of the

signal. This was necessary to ensure that manipulations of spectral slope only affected

the periodic energy for each stimulus, while leaving the aspiration noise of each stimulus

constant and unchanged. To achieve this, two versions of each base stimulus were










synthesized. One version was synthesized by setting AH (amplitude of aspiration) to 0 dB

and AV (amplitude of voicing) to 60 dB. This resulted in the synthesis of a vowel with no

aspiration noise. Furthermore, OQ (open quotient) was set to 30% and TL (tilt) was set to

15%. The second version of the same vowel was generated by setting the AH to 50 dB

but setting AV to 0 dB. This resulted in a vowel with no harmonic energy, but one where

the formants were excited using the aspiration noise alone. This approach provided the

harmonic spectrum as well as the aspiration noise spectrum for each of the ten base

stimuli.

Table 2.1. Parameters used to generate the 10 vowel stimuli*.


FO
AV
OQ
SQ
TL
FL
AH
FNP
BNP
F1
B1
F2
B2
F3
B3
F4
B4
F5
B5


ML1
133.1
60
40
200
0
10
35
180
1000
661
200
1122
200
2281
300
4198
400
4415
500


ML2
113.7
60
55
200
10
10
40
180
1000
559
400
1214
200
2340
300
3383
400
4396
500


ML3
115.5
60
65
200
20
10
50
180
1000
732
600
1244
200
2497
300
3362
400
4533
500


ML4
117.0
60
75
200
30
10
60
180
1000
456
800
1187
150
2463
200
3405
250
4194
300


ML5
134.4
60
85
200
40
10
80
180
1000
814
1000
1473
200
2250
250
3701
300
4990
350


FM1
220.4
60
40
200
0
10
35
180
1000
891
200
1587
200
3083
300
3870
400
4761
500


FM2
209.0
60
55
150
10
10
40
180
1000
759
400
1333
200
2930
300
4232
400
4736
500


FM3
209.1
60
65
350
20
10
50
180
40
1050
600
1470
200
3000
300
4000
400
4990
500


FM4
195.5
60
75
200
30
10
60
280
90
977
800
1326
150
2905
200
4651
250
4990
300


FM5
200.7
60
85
200
40
10
80
180
30
957
1000
1619
200
2877
250
4274
300
4883
350


ML refers to male synthetic voices and FM refers to female synthetic voices. The
abbreviations on the left hand side of the table refer to the acoustic parameters in each
stimulus and are standard parameters found in a Klatt-synthesizer. All abbreviations are
shown in the Appendix.

A series of low-pass finite impulse response 2 (FIR2) filter were generated in

MATLAB 7.1 (The MathWorks Inc., 2004) to manipulate the spectral slope of the









periodic energy for the ten base stimuli. FIR2 low-pass filters were used because they

allow manipulation of the spectral slope of a signal without affecting the other parameters

of the signal. These filters were generated with a maximum attenuation at cutoff

frequency of 1 dB, and a minimum attenuation at a high frequency of 120 dB. Each of

the ten stimuli was manipulated using these filters to generate two 10-step continue

varying in their spectral slope. The stimuli in each of these two continue varied in terms

of their spectral slope in increments of 3 dB/octave, ranging from -3 dB/octave to -30

dB/octave. The first continuum included stimuli that were filtered with a fixed- or

constant cutoff frequency of 500 Hz. This condition is henceforth referred to as CC

(constant cutoff). This condition ensured that the spectral slope for all stimuli was

manipulated around at fixed cut-off frequency. The 500 Hz cut-off was selected so that

the first formant for all stimuli was above the cut-off frequency. However, a fixed cut-off

frequency affected male and female stimuli differently. Male stimuli, with a lower

fundamental frequency, had a greater number of harmonics below 500 Hz as compared to

the female stimuli which had a higher fundamental frequency. If the total energy in the

low frequency region or the harmonic relationships for the first few harmonics played a

role in cueing breathiness, such differences in stimuli may affect the final results. To

further investigate this possibility, a second continuum of stimuli was generated. This

continuum was generated with a cutoff frequency between the second harmonic (H2) and

the third harmonic (H3) of each base synthetic voiced stimuli to account for the

differences between the ranges of the average fundamental frequencies according to

gender. This condition was labeled VC (varying cutoff). The amplitude of the first

harmonic H1 has been found to be correlated with breathiness in past research (Huffman,









1987). Therefore, the second stimulus continuum resulted in a series of stimuli that

varied in their slope, but had the same number of harmonics below the filter cut-off

frequency and had a constant H1 amplitude. A total of 200 stimuli were thus generated

(10 base stimuli X 2 continue X 10 stimuli/continua).

The aspiration noise for each of the ten base stimuli was then added to the two

hundred stimuli in the CC and VC continue. However, two additional steps needed to be

performed before adding the aspiration noise. First, the aspiration noise for each voice

was appropriately amplified to obtain a constant signal-to-noise ratio (SNR) of 25 dB,

using MATLAB 7.1. This was essential to create a proper balance between the periodic

signal and the aspiration noise, so that neither of these aspects overpowered the effects of

the other. An SNR of 25 dB was chosen based on pilot experiments that showed this

SNR to be ideal for the present experiment. Pilot experiment found that an average SNR

of 25 dB resulted in stimuli where listeners were still able to detect differences in the

voiced signal for each base stimulus. The accuracy of the algorithm used for equating the

SNR was further verified by calculating the SNR for these stimuli using a commercially

available software package, TF32 (Milenkovic, 1997). The SNR for all stimuli were

calculated using TF32 and were found to range from 22.8 dB to 25.5 dB. However, the

SNR in a single stimulus continuum was found to vary within a range of +/- 0.9 dB.

Second, the aspiration noise for each voice was temporally shifted to align it with

the filtered periodic signal. The filtering for both CC and VC conditions created a 257

point shift at the beginning of the voiced signal for each stimulus. In order to properly

add the noise back into the voiced signal for each stimulus, a 257 point shift was added to

the beginning of the ten base noise signals. Once this was accomplished, the ten base









noise signals were added back to the corresponding set of voiced stimuli. Both of these

steps were accomplished in MATLAB 7.1. Next, the 257 point delay at the beginning of

each new stimulus (filtered periodic signal + amplified and temporally shifted aspiration

noise) was removed by deleting the zeros at the beginning of the signal using Adobe

Audition 1.0 (Adobe Systems Inc., 2003). Lastly, the stimuli were resampled so as to

have a sampling frequency of 24,414 Hz. This was necessary to match the clock speed of

the A/D hardware used to present the stimuli to listeners (RP2, Tucker Davis Technology

Inc., 2000).

Perceptual Ratings

The perceptual experiment was divided into two sessions, with listeners rating 10

randomized lists of stimuli per session. Each session lasted approximately 45 minutes.

Listeners were seated in a sound treated booth, approximately 7 ft (L) X 7 ft (B) X 6 ft

(H). The stimuli were presented through an RP2 processor monoaurally through the right

ear using ER-2 ear inserts (Etymotic Inc.) at an intensity of 80 dB SPL. Monoaural

presentations were preferred to enable comparison of results with other experiments that

have used an auditory processing front-end to study breathiness (Shrivastav et al., 2003).

Binaural integration of auditory signals complicates certain steps in most auditory

models, hence it was avoided. There is no evidence to believe that perceptual judgments

for breathiness may vary between monoaural and binaural presentation in normal hearing

listeners.

Listeners were asked to rate the level of breathiness of each voice stimulus using a

seven-point rating scale, where a value of 1 indicated minimum breathiness and a value

of 7 indicated maximum breathiness. Listeners were instructed to rate only the

breathiness of each stimulus presented and to avoid making judgments based on pitch or









loudness. No definition of breathiness was provided. Ratings were made using a

computer monitor and a keyboard.

Twenty randomized lists of voice stimuli (10 CC voice stimuli and 10 VC voice

stimuli) were prepared and organized in SykofizX 2.0 software (Tucker Davis

Technologies Inc., 2005). Within each list, each of the ten stimuli was presented five

times in random order, for a total of 50 stimuli per list. Research has shown that

averaging multiple ratings of each stimulus provides a more accurate measure of a

listener's perception of voice quality (Shrivastav, Sapienza & Nandur, 2005). Each

stimulus was 489 ms in duration. Eleven milliseconds were removed from the original

signal, consisting of 500 ms, when the zeros were removed from the 257 point shift

added at the beginning of each signal in MATLAB 7.1. Listeners were provided a

maximum of 8 seconds to make their response before being presented with the next

stimulus. A short break (approximately 2-3 minutes) was provided between every 3-4

lists to minimize fatigue. The five ratings obtained for each voice stimulus from each

listener were averaged to obtain a single rating. These ratings were then averaged to

obtain a group mean rating for each voice stimulus.

Statistical Analyses

Intra- and inter-judge reliability was determined using Pearson's correlation

coefficient for both, CC and VC series. Intra-judge reliability was measured by

determining the average correlation between each of the five ratings for each stimulus

made by each listener. Inter-judge reliability was measured by determining the

correlation between each listeners mean rating for each stimulus.

A linear regression analysis was performed in SPSS 11.0 (SPSS Inc., 2002). This

was used to model the relationship between listener's mean breathiness ratings









(dependent variable) and spectral slope variation (independent variable) for both CC and

VC series. A regression function containing the y-intercept and slope for each series was

created. The variance and R-square values for each series was also calculated.

A two-way analysis of variance (ANOVA) was also performed as a confirmatory

test to determine if the mean breathiness ratings for stimuli at the two ends of the spectral

slope continuum (-3 dB/octave and -30 dB/octave) in each stimulus set in both CC and

VC series were significantly different from each other. The ANOVA was also used to

determine if any effects of gender (male vs. female stimuli) were observed. Mean

breathiness ratings served as the dependent variable whereas spectral slope and gender

served as the two independent variables. Any interaction between the two independent

variables was also investigated.

Acoustical Analyses

The ten base stimuli containing only the harmonic energy (AH = 0; AV = 60) were

further analyzed to determine some of their acoustic characteristics of the stimuli. This

was necessary to determine differences in the perceptual judgments across stimulus

series. First, the difference between the amplitudes of the first and second harmonics was

calculated (H1 H2). This was done because past research has indicated that first

harmonic dominance may play an important role in cueing breathiness (Huffman, 1987;

Hanson, 1997). This would also help explain whether the harmonic energy signals

differed in the low frequency region across stimuli. The intensity of the fundamental and

the second harmonics were corrected for the effects of the formant frequency using the

formula described by Hanson (1997). This correction allows a more direct comparison of

H1 H2 across stimuli varying in their fundamental and formant frequencies. The

corrected H1 H2 is indicated by HI* H2* and is calculated as follows:









H1 -H2* = (H1 K) (H2 K)

where,

H1 = Amplitude of the first harmonic,

H2 = Amplitude of the second harmonic, and,

K = Correction factor.

The value ofK is given by the formula:


K = 20 x log io( F2-[ 22]2


where,

Fl = Frequency of the first formant, and,

f = Frequency where the harmonic is located.

Another measurement to study differences across stimulus series included the

calculation of total power in the high frequency region for stimuli with no aspiration

noise. This was done because the ten base stimuli differed in the overall acoustic

characteristics (for example, differences in formant frequencies, formant bandwidths,

harmonic density, etc.) and these affect the total energy in higher frequencies. To make

these calculations, the stimuli were first normalized for overall power and then filtered

using a band-pass Butterworth filter. This band-pass filter had cut-off frequencies of 1500

Hz and 5000 Hz, transition bands of 100 Hz and a stop attenuation of 75 dB and was

generated using Adobe Audition 1.0. All base stimuli with no aspiration noise (AV = 60

dB, AH = 0 dB) were filtered and the total RMS power of the filtered signals was

calculated.

Finally, the characteristics of the aspiration noise for each stimulus series were

analyzed. This was done because even though the noise signals at source were held









constant and were not manipulated in this experiment, the various vocal tract

configurations for each voice stimuli would be different and influence the formants for

each voice. The noise characteristics were determined by studying the signals generated

by the synthesizer with the amplitude of voicing set to zero and amplitude of aspiration

noise set to 50 dB (AV = 0 dB, AH = 50 dB). These signals were first normalized for

average RMS power and were then analyzed using the software TF32. To describe the

nature of the noise spectrum, it was characterized as a standard probability distribution

function and its first four moments (mean, standard deviation, skewness, and kurtosis)

were calculated. Differences in these moments can be used to describe the differences in

the overall shape of the noise spectra. The procedure used for this analysis was based on

Forrest, Weismer, Milenkovic, and Dougall (1988).















CHAPTER 4
RESULTS

Listener Reliability

The intra-judge reliability for each listener was determined using Pearson's

correlation to examine the relationship between each of the five ratings provided by the

listeners. Separate analyses were performed for the CC and VC series. For the CC series,

the mean correlation for the ten listeners was 0.69 with a range of 0.31 to 0.97. The mean

correlation in the VC series was 0.71 and ranged from 0.33 to 0.97. These indicate a

moderately significant correlation. Table 3.1 lists the intra-judge reliability for each

listener in the CC and VC series.

Table 3.1. Intra-rater reliability for the CC and VC series
Pearson's r
Listener CC VC
L1 0.31 0.33
L2 0.95 0.97
L3 0.83 0.87
L4 0.95 0.96
L5 0.69 0.58
L6 0.80 0.94
L7 0.43 0.54
L8 0.81 0.83
L9 0.97 0.92
L10 0.31 0.37
Mean 0.69 0.71


The inter-judge reliability was determined by calculating the Pearson's correlations

between each listener's average ratings. The mean inter-judge reliability for the CC series

was 0.47 with a range of -0.23 to 0.91. Table 3.2 lists the inter-judge reliability between

every listener for the CC series. For the VC series, the mean inter-judge reliability for the









ten listeners was 0.55 with a range of 0.11 to 0.91. The inter-judge reliability for each

listener in the VC series is presented in Table 3.3.

At first glance, the inter-judge reliability measures for both CC and VC series

appear rather low; however, as discussed later, listeners did not vary much in their

breathiness ratings across an increasing spectral slope per stimulus set. The low

correlation may reflect a lack of variation in perceived breathiness across stimuli, rather

than an inability of the listeners to rate the stimuli consistently.

Table 3.2. Inter-rater reliability for the CC series
L1 L2 L3 L4 L5 L6 L7 L8 L9 L10
L1 1 0.69 0.71 0.67 0.48 0.64 0.13 0.44 0.30 0.04
L2 1 0.87 0.91 0.62 0.85 0.05 0.66 0.63 0.39
L3 1 0.81 0.69 0.86 0.27 0.58 0.36 0.29
L4 1 0.50 0.76 0.15 0.70 0.45 0.33
L5 1 0.80 0.18 0.53 0.44 0.16
L6 1 0.21 0.68 0.46 0.35
L7 1 0.34 -0.23 0.03
L8 1 0.64 0.35
L9 1 0.35
L10 1

Table 3.3. Inter-rater reliability for the VC series
L1 L2 L3 L4 L5 L6 L7 L8 L9 L10
L1 1 0.46 0.45 0.35 0.51 0.41 0.30 0.37 0.22 0.21
L2 1 0.81 0.84 0.81 0.91 0.60 0.79 0.35 0.68
L3 1 0.53 0.78 0.80 0.73 0.63 0.46 0.73
L4 1 0.62 0.73 0.21 0.66 0.11 0.54
L5 1 0.83 0.63 0.60 0.43 0.69
L6 1 0.61 0.67 0.34 0.65
L7 1 0.62 0.55 0.52
L8 1 0.24 0.51
L9 1 0.41
L10 1


Effects of Spectral Slope on Breathiness Ratings

For the CC series, the overall mean rating for the male voices was 5.7 with a

standard deviation of 0.14, while the mean rating for the female voices was 3.7 with a











standard deviation of 0.5. Figure 3.1 shows the listener mean ratings for the male voices


and Figure 3.2 shows the listener mean ratings for the female voices in the CC series


along a continuum of increasing spectral slope. For the VC series, the mean rating for the


male voices was 5.7 with a standard deviation of 0.36, while the mean rating for the


female voices was 3.5 with a standard deviation of 0.30. Figure 3.3 shows the mean


ratings for the male voices and Figure 3.4 shows the mean ratings for the female voices in


the VC series along a continuum of increasing spectral slope.


7



--MALE1
5 i- L i L i
MALE2
S44 ---------- MALE3
'MALE4
3 ---------
3-
MALE5
G2


-3 -6 -9 -12 -15 -18 -21 -24 -27 -30
Spectral slope variation (dBloctave)

Figure 3.1. Mean breathiness ratings for the male speakers in the CC series


7

6
S--FEML1
5 - - -
FEML2
E 4 ...... FEML3
S- FEML4
3 FEML5

1 I I-I-I- I- I I I-
2


-3 -6 -9 -12 -15 -18 -21 -24 -27 -30
Spectral slope variation (dBloctave)

Figure 3.2. Mean breathiness ratings for the female speakers in the CC series















c

-5
,m


4
3

|2


- MALE1
- MALE2
MALE3
MALE4
- MALE


-3 -6 -9 -12 -15 -18 -21 -24 -27 -30
Spectral slope variation (in dB)


Figure 3.3. Mean breathiness ratings for the male speakers in the VC series





7


6

5

S4

3
2-
I-


- FEML1
- FEML2
- FEML3
- FEML4
- FEML5


-3 -6 -9 -12 -15 -18 -21 -24 -27 -30
Spectral slope variation (in dB)



Figure 3.4. Mean breathiness ratings for the female speakers in the VC series




Table 3.4 lists the overall listener mean ratings and standard deviation according to


spectral slope variation for both the CC and VC series. The difference between the lowest


and the highest mean ratings for the CC series is 0.32. The VC series demonstrates a


difference of 0.48 between the lowest and the highest mean ratings. These differences


~L

51~C~-r 1 I L


- -









are rather low, considering that a large change (-3 dB/octave to -30 dB/octave) was made

in each stimulus continuum.

Table 3.4. Overall listener mean ratings and standard deviation with increasing spectral
slope
CC VC
slope Mean SD Mean SD
(dB/octave)
1 4.46 1.27 4.24 1.38
2 4.50 1.24 4.38 1.29
3 4.65 1.13 4.48 1.25
4 4.65 1.13 4.58 1.23
5 4.71 1.07 4.63 1.16
6 4.76 1.11 4.67 1.17
7 4.73 1.04 4.72 1.10
8 4.77 1.08 4.67 1.15
9 4.78 1.06 4.65 1.19
10 4.73 1.06 4.70 1.17


A regression analysis was used to model the relationship between listeners' mean

breathiness ratings (dependent variable) and spectral slope variation (independent

variable) in each series. A linear regression was performed to predict the listener mean

ratings from spectral slope for both CC and VC series. For the CC series, the effects of

listener mean ratings predicted by the following regression function accounted for 73.9%

of the variance in the perceptual ratings (R-square = 0.739):

Breathiness Rating = 4.493 0.011 spectral slope

For the VC series, the effects of listener mean ratings were predicted by the equation

listed below:

Breathiness Rating = 4.323 0.015 spectral slope

This equation accounted for 74.4% of the variance in the mean ratings (R-square =

0.744). Figure 3.5 and Figure 3.6 demonstrate the relationship between listeners' mean

breathiness ratings and spectral slope variation for both CC and VC series.

















6.0



5.0


U)
g' 4.0

rY

) 3.0
c
(-
2.0
C
03
r 1.0


-24 -21 -18 -15 -12


Spectral Slope (dB/octave)


Figure 3.5. Relationship between listeners'
variation for the CC series.




7.0



6.0'



5.0*


-30 -27 -24 -21 -18 -15 -12 -9 -6


mean breathiness ratings and spectral slope


Rsq = 0.7438


Spectral Slope (dB/octave)



Figure 3.6. Relationship between listeners' mean breathiness ratings and spectral slope
variation for the VC series.


7.0 -


p -


30 -27


-9 -6


Rsq = 0.7386


p









As a confirmatory test, a two-way analysis of variance (ANOVA) was performed to

determine if the mean breathiness ratings for stimuli at the two ends of the continuum (-3

dB/octave and -30 dB/octave) were significantly different from each other. Perceptual

ratings of breathiness served as the dependent variable whereas spectral slope (-3

dB/octave or -30 dB/octave) and gender (male or female) served as the two independent

variables. Any interaction between the two independent variables was also investigated.

For the CC series, no significant effects of spectral slope were observed on

breathiness ratings (F() = 2.719; p = 0.119). This further supports the poor correlation

between spectral slope breathiness ratings in the CC series. However, a significant main

effect for gender was obtained showing that the male voices were perceived to be

significantly more breathy than the female voices (F(I)= 159.191; p < 0.001). No

significant interaction between spectral slope and gender was observed (F() = 0.782; p =

0.390).

A significant main effect of spectral slope on breathiness ratings was demonstrated

in the VC series (F(1)= 8.236; p = 0.011). This supports the slightly larger difference

listeners were able to perceive between the lowest and the highest mean breathiness

ratings in this series as compared to the CC series. A significant main effect of gender

was obtained demonstrating that the male voices were perceived to be significantly

breathier than the female voices (F() = 202.093; p < 0.001). No significant interaction

between spectral slope and gender was observed (F() = 1.153; p = 0.299).

Acoustic Analyses

The difference between the first harmonic amplitude (H1) and the second harmonic

amplitude (H2) were measured in the ten base stimuli containing only harmonic energy

(AH = 0 dB; AV = 60 dB) for both the CC and VC series. This was analyzed to









determine if the difference between the amplitude of H1 and H2 had any effects on

listeners' perception of breathiness in the stimuli presented to them. H1 and H2 were not

directly controlled in this experiment since these were always below the filter cut-off

frequency. The difference in amplitude between H1 and H2 is an indication of the open

quotient in a signal (Hanson, 1997) and open quotient / amplitude of Hi have been

indicated as a predictor of breathiness in several studies (Klatt & Klatt, 1990;

Hillenbrand, Cleveland, & Houde, 1994; Hillenbrand & Houde, 1996).

For the CC series, the male stimuli demonstrated an HI* H2* mean of -2.8 dB,

with a standard deviation of 1.21 dB. The female stimuli in this series demonstrated an

HI* H2* mean of -4.72 dB, with a standard deviation of 2.93 dB. For the VC series,

the male stimuli demonstrated an H H2* mean of-1.88 dB, with a standard deviation

of 1.23 dB. The female stimuli in this series demonstrated an HI* H2* mean of -4.86

dB, with a standard deviation of 2.87 dB. Therefore, these results indicate that on

average male stimuli had a more dominant H1 amplitude than the female stimuli. Table

3.5 lists the H H2* with their corresponding mean ratings and standard deviations for

each voice in the CC and VC series.

Table 3.5. Relationship between HI* H2* and mean rating for each stimuli in both CC
and VC series
CC VC
H1* H2* (in dB) Mean Rating HI* H2* (in dB) Mean Rating
MALE1 -4.9 5.78 -2.9 5.47
MALE2 -2.7 5.82 -2.7 5.79
MALE3 -2.4 5.53 0.2 6.19
MALE4 -2.1 5.68 -2.1 5.76
MALES -1.9 5.60 -1.9 5.18
FEML1 -9.1 3.15 -9.2 3.15
FEML2 -6.1 4.04 -6.2 3.68
FEML3 -2.2 3.34 -2.6 3.33
FEML4 -2.2 4.20 -2.3 3.71
FEML5 -4 3.57 -4 3.46









The total RMS power in the high frequency region for the ten base harmonic

energy stimuli (AH = 0 dB; AV = 60 dB) was calculated to determine if any further

differences were found across stimulus sets for both series. This analysis revealed that

male voices had a mean total RMS power of -64.88 dB (SD = 8.18 dB) between 1500

and 5000 Hz, whereas female voices had a mean total RMS power of -41.14 dB (SD =

8.62 dB) in this same frequency range. There is a difference of -23.74 dB between the

mean RMS power in male voices and female voices. The results indicate that the voicing

source for the female stimuli resulted in greater power in the high frequency band than

seen in male speakers. Table 3.6 lists the total RMS power for each of the ten base

harmonic energy stimuli. Figure 3.7 demonstrates how the spectra for the male stimuli

(e.g., MALE4) and the female stimuli (e.g., FEML5) differ in the higher frequency

region. This figure shows that male speakers tend to have very little harmonic energy

above 2800 Hz, whereas female speakers had harmonic energy up to 5000 Hz.

Table 3.6. Total RMS power and mean ratings for ten base harmonic signal stimuli
Total RMS Power
Stimulus FO Total RMS Power Mean Rating (CC) Mean Rating (VC)
(dB)
MALE1 132 -71.41 5.78 5.47
MALE2 114 -69.99 5.82 5.79
MALE3 116 -63.26 5.53 6.19
MALE4 117 -68.42 5.68 5.76
MALES 135 -51.32 5.60 5.18
FEML1 220 -35.01 3.15 3.15
FEML2 209 -54.58 4.04 3.68
FEML3 209 -38.16 3.34 3.33
FEML4 196 -44.49 4.20 3.71
FEML5 200 -33.47 3.57 3.46











120
100 ---------------------------------------


0 80 -
O FEML5
60 -----------------
S60- MALE4
E 40 -

20


0 2500 5000 7500 10000
Frequency (Hz)


Figure 3.7. Example of gender differences in the power spectrum

The spectral characteristics of the noise were further analyzed in the ten base noise

stimuli (AH = 50 dB; AV = 0 dB). Each noise spectrum was treated as a probability

distribution function and its first four moments were calculated (mean, SD, skewness and

kurtosis). These analyses were completed using TF32 (Milenkovic, 1997). In the male

stimuli, the noise spectra had a mean of 1647 Hz, a standard deviation of 1220 Hz,

skewness of 4.01, and a kurtosis of 30.06. The female stimuli demonstrated a mean

frequency of 1826 Hz, a standard deviation of 920 Hz, skewness of 2.92, and a kurtosis

of 17.43. These results indicate that the male stimuli used in this experiment had a lower

average noise frequency than for female stimuli. The aspiration noise in the male stimuli

was also observed to be more skewed to the right than the female stimuli. Finally, the

male stimuli were observed to have larger kurtosis than the female stimuli. Table 3.7 lists

the spectral moments for each of the ten base noise stimuli.









Table 3.7. Spectral moments for ten base noise signal stimuli
Mean (Hz) SD (Hz) Skew Kurtosis
MALE1 1006 1142 5.191 36.03
MALE2 1415 1458 3.517 18.294
MALE3 1876 1178 2.958 19.472
MALE4 2055 1595 2.552 11.611
MALES 1882 729 5.848 64.898
FEML1 2033 997 1.791 9.347
FEML2 1538 886 3.572 23.256
FEML3 1763 824 3.471 24.844
FEML4 1672 976 3.293 16
FEML5 2126 927 2.457 13.678


Summary of Results

Listeners demonstrated a moderately significant intra-judge reliability in both

series. However, these listeners demonstrated a weaker inter-judge correlation in both

CC and VC series. This may be due to the fact that an increase in spectral slope had little

effect on listeners' perception of breathiness. The difference between the lowest and the

highest mean breathiness ratings across spectral slope were relatively small, even though

the VC series demonstrated a slightly greater and statistically significant difference

between the two end-points of the continuum. A regression analysis supported this

finding and showed a weak relationship between the spectral slope and breathiness

ratings.

Male stimuli were rated significantly higher in terms of breathiness than the

female stimuli for both series. Acoustic analyses of the stimuli showed that the male and

female stimuli also differed in their H H2*, the average power of the harmonics in the

high frequency region and in the spectral characteristics of their aspiration noise. These

differences may be responsible for the gender effect found in this experiment.














CHAPTER 5
DISCUSSION

The goal of the present study was to determine the effects of changes in spectral

slope on the perception of breathiness. This was done because the role of spectral slope

on perceived breathiness remains unclear, with some studies indicating that spectral slope

plays an important role in the perception of breathiness (Huffman, 1987; Klatt & Klatt,

1990; Childers & Ahn, 1995), while other studies such as Hillenbrand (1988), stated that

spectral slope was not associated with breathiness. The results of this present study are

discussed below.

Reliability measurements were taken to determine the consistency of listeners

within themselves and with one another in making perceptual ratings. Pearson's

correlation revealed that the intra-judge reliability varied among listeners in both the CC

and VC series. The CC series demonstrated only a moderate level of intra-judge

reliability (0.69). Three listeners had intra-judge reliability under 0.50. The VC series

also demonstrated a moderate level of intra-judge reliability (0.71). Two listeners had

intra-judge reliability under 0.50. The fact that listeners were not able to perceive much

of a difference in levels of breathiness in the stimuli presented to them may be a reason

why they demonstrated moderately high levels of reliability. In order to obtain a high

correlation between two variables, there must be sufficient variability in the data. If there

is no variation, then the two variables will not demonstrate high levels of correlation.

The inter-judge reliability also varied among listeners in both the CC and VC

series. Both, CC and VC series demonstrated an overall moderate level of inter-judge









reliability (Pearson's correlation of 0.47 and 0.55, respectively). Although these measures

appear rather low, this may again reflect the small variance in the perceptual data.

Therefore, the low inter-judge reliability likely results from the nature of the stimuli

rather than differences across listeners. This was further confirmed by the findings

discussed below.

Perceptual ratings indicated that there is little change in perceived breathiness when

spectral slope is manipulated in both CC and VC conditions. The difference between the

lowest and the highest mean breathiness ratings across spectral slope position in the CC

and VC series demonstrated a difference of 0.32 and 0.48, respectively. Although these

differences were relatively small; the mean rating for the stimuli located at -3 dB/octave

and at -30 dB/octave in the VC series were found to be statistically significant, according

to a two-way analysis of variance (ANOVA). A linear regression analysis examined the

relationship between spectral slope variation and listeners' mean breathiness ratings in

both series. For both CC and VC series, the regression function accounted for a large

amount of variance in the perceptual data (R-squares of 0.739 for the CC series and 0.744

for the VC series). However, the slope of these regression functions were very low (-

0.011 and -0.015 for the CC and VC series, respectively) suggesting that variations in

spectral slope had only a small effect on perceived breathiness in these stimuli.

Although these results agree with some past research (for example, Hillenbrand,

1988), these contradict the findings of some other studies that have found measures of

spectral slope to correlate with breathiness (Huffman, 1987; Klatt & Klatt, 1990; Childers

& Ahn, 1995; Bhuta, Patrick, & Garnett, 2004). These differences may be attributed to

certain methodological differences. The current study systematically manipulated spectral









slope in a controlled manner. Unlike previous studies that used natural stimuli, factors

such as SNR, open quotient, and first harmonic amplitude were controlled in this current

study to minimize their influence on the results. These factors have been shown to be

predictors of breathiness in several studies (Huffman, 1987; Klatt & Klatt, 1990;

Hillenbrand, Cleveland, & Erickson, 1994; Childers & Ahn, 1995; de Krom, 1995;

Martin, Fitch, & Wolfe, 1995; Hillenbrand & Houde, 1996; Bhuta, Patrick, & Garnett,

2004; Shrivastav & Pinero, 2005) and these may have co-varied with changes in spectral

slope. The SNR for all stimuli was held constant at 25 dB and the open quotient was set

to 30% for every stimulus. Additionally, the SNR value of 25 dB may also partly explain

why spectral slope variation did not affect breathiness in the present experiment. This is

further discussed below.

The findings of the present experiment may also be explained using the partial

loudness model described by Shrivastav and Sapienza (2003). Since partial loudness is

related to the level of the harmonic energy relative to that of the aspiration noise, changes

in either of these parameters can affect partial loudness. The stimuli used in this

experiment varied in their spectral slope, but had a constant SNR, obtained by modifying

the overall level of the harmonic energy while keeping a constant aspiration noise level.

An increase in the spectral slope without any changes to the level of aspiration

noise would result in a decrease in partial loudness of the harmonic signal. The partial

loudness is also dependent on the spectral shape of the signal and the masker. Therefore,

once the aspiration noise completely masks the harmonic signal at specific frequencies, a

further change in spectral slope would have little affect on partial loudness. The results of

this study follow this pattern in that on average, listeners are able to detect differences in









breathiness in a stimulus among the first two instances of an increasing spectral slope in

the CC series and among the first three instances of an increasing spectral slope in the VC

series. Presumably, an increase in spectral slope after these levels provides no additional

masking. Thus, there is no further change in partial loudness, or in perceived breathiness.

The fact that the spectral slope variation resulted in a slightly greater increase in

breathiness for the VC series may be related to the lower filter cut-off frequency in these

series (particularly for the male stimuli). A lower filter cut-off frequency may affect

partial loudness to a greater degree because the filtering would affect the level of the

harmonic signal to a greater degree.

This model would further predict that changes in spectral slope may have failed to

affect the breathiness for these stimuli because the SNR of 25 dB may have already

masked the harmonic energy significantly. A further increase in spectral slope may not

have resulted in any significant change in partial loudness of the harmonic energy. This

model would further predict that if the SNR were increased, a change in spectral slope

would result in a greater change in breathiness. This is because a higher SNR would

result in a greater difference between the levels of the harmonics and the aspiration noise.

A change in spectral slope for these stimuli would lead to a greater change in masking,

and hence partial loudness and breathiness. However, this prediction needs to be

empirically tested.

A significant gender effect was also observed for the mean ratings of breathiness.

As shown in Figure 3.1 and 3.2, the five male synthetic voices were rated to be more

breathy (ratings between 5.2 and 5.8) than female voices (ratings between 2.7 and 4.4).

Figures 3.3 and 3.4 demonstrate similar differences for the VC series. A two-way









analysis of variance (ANOVA) confirmed the gender differences as being significant. It

is interesting that the synthetic male voices were perceived to be breathier than the

synthetic female voices, since female voices have been reported to be breathier voice

quality than male voices (Colton & Casper, 1995).

Closer examination of the acoustic properties of the harmonic signals in these

stimuli demonstrated several differences between the male and female stimuli. First, male

stimuli had a more dominant H1 amplitude than the female stimuli. Second, calculation

of total RMS power in specific frequency bands revealed that the female stimuli had

greater harmonic energy between 1500 Hz and 5000 Hz as compared to the male stimuli.

Upon examining the range of the last harmonic in the male and female stimuli, it was

noted that the last harmonic in the male stimuli occurred between 1000 Hz and 1500 Hz,

while the last harmonic in the female stimuli occurred between 1700 Hz and 2400 Hz.

This goes along with the fact that males have larger vocal tracts than females, resulting in

lower resonant frequencies and lower formant peaks, which in turn affect the harmonic

and noise signals of a stimulus. Third, the aspiration noise spectra for the male and

female stimuli differed in several ways. The male stimuli demonstrated a lower mean

frequency than the female stimuli. The male stimuli also demonstrated a greater skewness

to the right and had a greater level of kurtosis than the female stimuli. Together, these

differences in the harmonic and aspiration noise spectra leads to a greater influence of

noise in the male stimuli, as compared to the female stimuli.

The kurtosis of one male stimuli (MALE5) was almost three times as large as the

next highest stimuli. This voice stimulus may have this large amount of kurtosis due to

its noise stimuli occurring at a low level. If this stimulus is removed, the overall









difference between male and female stimuli is not very significant. These acoustic

differences in the harmonic energy and aspiration noise between the male and female

stimuli directly affect the partial loudness patterns for the voices and can explain the

gender differences observed in the perceptual ratings.

The results of this experiment must be interpreted in light of the fact that: (1) the

cutoff frequency was set to 500 Hz or between H2 and H3 of a stimulus; (2) the open

quotient was set to 30%; and (3) the SNR was set to 25 dB. If the three variables of

cutoff frequency, open quotient, and SNR are varied from the parameters used in this

study, the results may differ. For example, two cutoff frequencies used in this current

study yielded slightly different results in that the VC series demonstrated a slightly larger

range of perceptual ratings compared to the CC series. On the other hand, raising the

open quotient to a higher percentage would increase the amount of time the vocal folds

are open relative to the total duration of the period, thus increasing the H1 amplitude.

Lastly, decreasing the SNR would lead to a stimuli containing more noise than signal,

leading the noise aspect to dominate the harmonic energy. The effects of each of these

three factors needs to be empirically studied to obtain a complete understanding of how

spectral slope may affect breathiness.

A second limitation deals with the fact that the noise signal was kept constant for

all stimuli. This creates a problem, as was discussed in terms of the partial loudness

model. As spectral slope is increased, the same amount of noise could result in greater

masking of the harmonic energy. However, if the harmonic levels are too low, an

increase in the SNR will have no further affect on masking the harmonic energy. The

steeper spectral slopes in this current study may have been perceived as being breathier if









the SNR was maintained at a higher level. Future studies should test this possibility, as it

will help shed light on the appropriateness of partial loudness in predicting breathiness.

Another limitation deals with the use of synthetic stimuli. The synthetic stimuli

used in this experiment only had energy up to 5000 Hz. However, natural voices may

have energy (especially the aspiration noise) extending above this range. This loss of

high frequency energy in the synthetic stimuli may lead to somewhat different results as

compared to natural voices. This may further affect the perceptual ratings of breathiness.

Future experiments may need to consider the role of frequencies above 5 kHz in the

perception of breathiness.

The fact that only the vowel [a] was used in this study may also be considered a

further limitation of this study. Other vowels are produced with different vocal tract

configurations, which may lead to different outcomes. Connected speech has been shown

to produce some differing results when compared to vowels (Hillenbrand et al., 1996).

These considerations could be addressed in future studies.

Future studies should compare breathy voices found in healthy individuals with

breathy voices resulting from various voice disorders. The results of this current study

differ from those of previous studies that have found spectral slope to be a significant

predictor of breathiness (Huffman, 1987; Klatt & Klatt, 1990; Childers & Ahn, 1995).

One reason for these differences may be the choice of stimuli in these experiments. In

these studies, breathy voices found in healthy individuals were used to analyze various

measures of spectral slope, while this current study used voice stimuli consisting of a

variety of voice disorders. Both normal and disordered voices, consisting of various

levels of breathiness, should be examined in a future study under the same methodology.









It may be that breathy voices observed in healthy individuals has better SNR than found

in disordered voices. Examining this issue will help determine if the two groups of

voices are distinctly different or whether they constitute different regions on the same

continuum.

Future research should also verify the role of the other acoustic correlates

mentioned in previous studies. As mentioned previously, there are at least four different

acoustic cues related to breathiness. Some of these parameters are specific to only

breathiness, while others have been shown to be significant predictors of other voice

qualities. Many of these studies looked for correlations between an acoustic parameter

and the perception of breathiness without explicitly testing the effects of these parameters

on the perception of breathiness. These future studies should try to incorporate a

common theoretical framework that controls for every possible confounding variable,

which should lead to more accurate acoustic predictors of breathiness.

Once we are better able to know all of the predictors of breathiness, and other vocal

qualities for that matter, clinicians will be better able to objectively assess voice qualities

in individuals who present with a vocal pathology. Clinicians can then use these

measures as supplements to their subjective ratings of vocal qualities to gain a better

picture of a patient's voice condition. By obtaining objective measures, intra-rater and

inter-rater reliability measures will also improve, as objective measures would help yield

more consistent measures in measuring the clinical outcome in a patient over time and

also would add more consistency in communication across clinicians.














CHAPTER 6
CONCLUSIONS

The effects of spectral slope manipulations for voice stimuli were analyzed to

determine listeners' perception of breathiness. Two continue varying in spectral slope

were created. The stimuli in each continuum were filtered using high-pass filters with

slopes ranging from -3 dB/octave to -30 dB/octave in increments of 3 dB/octave. The

first continuum (CC series) contained stimuli which were low pass filtered at a constant

cutoff frequency of 500 Hz to ensure that the first formant of each stimulus would not be

filtered. The second continuum (VC series) contained stimuli which were filtered at a

cutoff frequency between H2 and H3 of each stimulus to ensure that each stimulus set

would have the same number of harmonics below the filter cutoff frequency.

Furthermore, the open quotient of each stimulus was set to 30% and the SNR was set at

25 dB.

Listeners' perceptual ratings demonstrated that as spectral slope was increased in

each set of stimuli there was little change in perceived breathiness for both CC and VC

series. This was confirmed statistically by performing a regression analysis, which

indicated a very low slope value between listeners' ratings from -3 dB/octave to -30

dB/octave for both series. A two-way ANOVA was also performed and indicated that

the mean breathiness ratings for the VC series demonstrated a small but significant

increase in the mean breathiness ratings for stimuli with the -30 dB/octave filter when

compared to the -3 dB/octave condition. No significant increase in breathiness was

observed for the CC series.









A significant gender effect for perceptual ratings of breathiness was also observed.

In both CC and VC series, the male stimuli were rated to be more breathy than the female

stimuli. This finding was confirmed statistically through a two-way ANOVA. The

acoustic properties of the harmonic signals in these stimuli revealed several differences

between the male and female stimuli with the male stimuli having greater H1 amplitude,

less harmonic energy in the higher frequency, and differences in the aspiration noise

spectra. Together, these differences may account for the differences observed in the

perceptual ratings between the male and female stimuli.

The effects of spectral slope variation as well as the gender differences obtained in

the present study may be explained on the basis of changes in the partial loudness of the

harmonic energy when it is masked by the aspiration noise. The small effect of spectral

slope variation may have resulted because of a relatively small SNR (25 dB). Based on

the partial loudness model, it is predicted that spectral slope variations would have a

greater effect on breathiness for a higher SNR. However, this needs to be empirically

verified.

In conclusion, this study indicates that spectral slope's role on the perception of

breathiness may be secondary to that of the aspiration noise. Unlike previous research

studies that found spectral slope to be important (Huffman, 1987; Klatt & Klatt, 1990;

Childers & Ahn, 1995), the present experiment found that spectral slope had a very small

effect on the perception of breathiness. The differences in these findings may relate to

differences in the other parameters for the stimuli (i.e., SNR, open quotient, first

harmonic amplitude, etc.) used in different experiments (Huffman, 1987; Eskenazi,

Childers, & Hicks, 1990; Klatt & Klatt, 1990; Hillenbrand, Cleveland, & Erickson, 1994;






46


Childers & Ahn, 1995; Martin, Fitch, & Wolfe, 1995; Hillenbrand & Houde, 1996;

Bhuta, Patrick, & Garnett, 2004; Shrivastav & Pinero, 2005). Future research should

investigate the effect of other such parameters in a systematic and controlled manner to

better understand their role on breathiness. This will result in the development of

appropriate models for voice quality perception as well as tools that will allow clinicians

to objectively assess individuals presenting with various levels of breathy vocal quality.














APPENDIX
DESCRIPTION OF PARAMETERS USED TO GENERATE TEN VOWEL STIMULI

Parameter MIN VAL MAX Description
FO 0 1000 5000 Fundamental frequency, in tenths of an Hz
AV 0 60 80 Amplitude of voicing, in dB
OQ 10 50 99 Open quotient (voicing open-time/period),
in %
SQ 100 200 500 Speed quotient (rise/fall time of open
period, LF model only), in %
TL 0 0 41 Extra tilt of voicing spectrum, dB down at 3
SkHz
FL 0 0 100 Flutter (random fluct inJO), in % of
maximum
AH 0 0 80 Amplitude of aspiration, in dB
FNP 180 280 500 Frequency of the nasal pole, in Hz
BNP 40 90 1000 Bandwidth of the nasal pole, in Hz
Fl 180 500 1300 Frequency of the first formant, in Hz
Bl 30 60 1000 Bandwidth of the first formant, in Hz
F2 550 1500 3000 Frequency of the second formant, in Hz
B2 40 90 1000 Bandwidth of the second formant, in Hz
F3 1200 2500 4800 Frequency of the third formant, in Hz
B3 60 150 1000 Bandwidth of the third formant, in Hz
F4 2400 3250 4990 Frequency of the fourth formant, in Hz
B4 100 200 1000 Bandwidth of the fourth formant, in Hz
F5 3000 3700 4990 Frequency of the fifth formant, in Hz
B5 100 200 1500 Bandwidth of the first formant, in Hz
* MIN represents the minimum value of the parameter. VAL represents the default value
which is applied if the user makes no changes. MAX represents the maximum value of
the parameter
**Table adapted from Klatt and Klatt (1990)















LIST OF REFERENCES


Bhuta, T., Patrick, L., & Garnett, J. D. (2004). Perceptual evaluation of voice quality and
its correlation with acoustic measurements. Journal of Voice, 18(3), 299-304.

Childers, D. G., & Ahn, C. (1995). Modeling the glottal volume-velocity waveform for
three voice types. Journal of the Acoustical Society ofAmerica, 97(1), 505-519.

Colton, R., & Casper, J. K. (1995). Understanding voice problems: A physiological
perspective for diagnosis and treatment. Baltimore: Williams and Wilkins.

de Krom, G. (1995). Some spectral correlates of pathological breathy and rough voice
quality for different types of vowel fragments. Journal of Speech andHearing
Research, 38, 794-811.

Eskenazi, L., Childers, D. G., & Hicks, D. M. (1990). Acoustic correlates of vocal
quality. Journal of Speech and Hearing Research, 33, 298-306.

Fairbanks, G. (1940). Voice and articulation drillbook. New York: Harper and
Brothers.

Fischer-Jorgensen, E. (1967). Phonetic analysis of breathy (murmured) vowels in
Gujarati. Indian Linguistics, 28, 71-139.

Forrest, K., Weismer, G., Milenkovic, P., & Dougall, R. N. (1988). Statistical analysis of
word-initial voiceless obstruents: Preliminary data.. Journal of the Acoustical
Society ofAmerica, 84(1), 115-123.

Gerratt, B. R., Kreiman, J., Antonanzas-Barroso, N., & Berke, G. S. (1993). Comparing
internal and external standards in voice quality judgments. Journal of Speech and
Hearing Research, 36, 14-20.

Hanson, H. (1997). Glottal characteristics of female speakers: Acoustic correlates.
Journal of the Acoustical Society ofAmerica, 101(1), 466-481.

Hillenbrand, J. (1988). Perception of aperiodicities in synthetically generated voices.
Journal of the Acoustical Society ofAmerica, 83(6), 2361-2371.

Hillenbrand, J., Cleveland, R. A., & Erickson, R. L. (1994). Acoustic correlates of
breathy vocal quality. Journal of Speech and Hearing Research, 37, 769-778.









Hillenbrand, J., & Houde, R. A. (1996). Acoustic correlates of breathy vocal quality:
Dysphonic voices and continuous speech. Journal of Speech and Hearing
Research, 39, 311-321.

Hirano, M. (1981). Clinical examination of voice. New York: Springer-Verlag.

Huffman, M. (1987). Measures of phonation type in Hmong. Journal of the Acoustical
Society ofAmerica, 81(2), 495-504.

Klatt, D., & Klatt, L. (1990). Analysis, synthesis, and perception of voice quality
variations among female and male talkers. Journal of the Acoustical Society of
America, 87(2), 820-857.

Klich, R. J. (1982). Relationships of vowel characteristics to listener ratings of
breathiness. Journal of Speech and Hearing Research, 25, 574-580.

Kreiman, J., & Gerratt, B. R. (1996). The perceptual structure of pathological voice
quality. Journal of the Acoustical Society ofAmerica, 100(3), 1787-1797.

Kreiman, J., & Gerratt, B. R. (1998). Validity of rating scale measures of voice quality.
Journal of the Acoustical Society ofAmerica, 104(3), 1598-1608.

Kreiman, J., & Gerratt, B. R. (2000a). Measuring voice quality. In R. D. Kent, & M. J.
Ball (Eds.), Voice quality measurement (pp. 73-101). San Diego, CA: Singular.

Kreiman, J., & Gerratt, B. R. (2000b). Sources of listener disagreement in voice quality
assessment. Journal of the Acoustical Society ofAmerica, 108(4), 1867-1876.

Kreiman, J., Gerratt, B. R., Kempster, G.B., Erman, A., & Berke, G.S. (1993).
Perceptual evaluation of voice quality: Review, tutorial, and a framework for
future research. Journal of Speech and Hearing Research, 36, 21-40.

Kreiman, J., Gerratt, B. R., Precoda, K. (1990). Listener experience and perception of
voice quality. Journal of Speech and Hearing Research, 33, 103-115.

Kreiman, J., Gerratt, B. R., Precoda, K., & Berke, G. S. (1992). Individual differences in
voice quality perception. Journal of Speech and Hearing Research, 35, 512-520.

Martin, D., Fitch, J., & Wolfe, V. (1995). Pathologic voice type and the acoustic
prediction of severity. Journal of Speech and Hearing Research, 38, 765-771.

Ostrem, J., & Fields, J. (2005). Tutorials: Voice production. Retrieved November 3,
2005, from The National Center for Voice and Speech Web site:
http://www.ncvs.org/ncvs/tutorials/voiceprod/tutorial/index.html.

Shrivastav, R., & Pinero, M. (2005). Effects of aspiration noise and spectral slope on
perceived breathiness in vowels. Journal of the Acoustical Society ofAmerica,
117(4), 2622-2623.






50


Shrivastav, R., & Sapienza, C. M. (2003). Objective measures of breathy voice quality
obtained using an auditory model. Journal ofAcoustical Society ofAmerica,
114(4), 2217-2224.

Shrivastav, R., Sapienza, C. M., & Nandur, V. (2005). Application of psychometric
theory to the measurement of voice quality using rating scales. Journal of Speech,
Language, and Hearing Research, 48, 1-13.

Wolfe, V., Cornell, R., & Palmer, C. (1991). Acoustic correlates of pathologic voice
types. Journal of Speech and Hearing Research, 34, 509-516.

Wolfe, V., & Martin, D. (1997). Acoustic correlates of dysphonia: Type and severity.
Journal of Communication Disorders, 30, 403-416.















BIOGRAPHICAL SKETCH

Mario Landera is a graduating master's student in the University of Florida

Department of Communication Sciences and Disorders. During his master's program, he

completed a master's thesis examining the effects of spectral slope on perceived

breathiness under the mentorship of Rahul Shrivastav, Ph.D., which was accepted as a

poster presentation at the 151st Acoustical Society of America (ASA) Meeting. Mr.

Landera received his B.S. in communication sciences and disorders from the Florida

State University in May 2004. In his senior year, he completed a senior honors thesis

examining social isolation in adolescents who stutter under the mentorship of Lisa Scott,

Ph.D., which was accepted as a poster presentation at the 2004 annual American Speech-

Language Hearing Association (ASHA) Convention. He was also recognized as the

outstanding senior in speech-language pathology during his senior year. Over his four

years of undergraduate studies, he was honored with membership into Phi Kappa Phi

honor society, Phi Sigma Theta honor society, Lambda Pi Eta honor society, and the

National Society of Collegiate Scholars. He has also been on the Dean's List for his

GPA throughout his college career.

Before beginning his graduate studies at the University of Florida, Mr. Landera was

accepted as a Board of Education fellow in the summer of 2004, where he was instructed

on the research process and writing. During his first year at the University of Florida as a

full-time graduate student, he worked as a graduate assistant at the Office of Graduate

Minority Programs, assisting in various recruitment and retention tasks targeting






52


underrepresented minority graduate students. In his second year as a graduate student at

the University of Florida, he worked as a graduate research assistant in the voice

perception lab in the Department of Communication Sciences and Disorders, under the

supervision of Rahul Shrivastav, Ph.D. His duties have included a review of literature on

voice quality, design of an experiment, generating appropriate stimuli, recruiting test

participants, and data collection and analysis. In July 2006, Mr. Landera will begin his

clinical fellowship year at the Miami Veteran's Affairs Medical Center in Miami, Florida.