Computational differences between whispered and non-whispered speech

Whispering is a common type of speech which is not often studied in speech technology. Perceptual and physiological studies show us that whispered speech is subtly di erent from phonated speech, and is surprisingly able to carry a tremendous amount of information. In this dissertation we con- sider the question: What makes whispering a good form of communication? We examine the di erences between normal phonated speech and whispered speech, and gauge the e ectiveness of state-of-the-art speech recognition al- gorithms at recognizing whisper. Our perceptual experiments add to the literature on the intelligibility of whispered speech. Comparisons with ASR results yield interesting insights into the di erences between the two systems. A method for building speech recognizers for whispered speech using lim- ited whispered speech data is proposed and evaluated. Our approach ef- fectively performs speaker-adaptation for whispered speech acoustic models without needing whispered speech from the target speaker. Results show im- provement over the standard speaker-independent models. Our work opens up additional avenues for research, which are outlined in the conclusions.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2011

Author dc:creator

• Lim, Boon Pang

Contributors dc:contributor

• Hasegawa-Johnson, Mark A.
• Sproat, Richard W.
• Shosted, Ryan K.
• Levinson, Stephen E.
• Lumetta, Steven S.

• speech recognition
• whispered speech
• modified rhyme test
• eigenvoices
• speaker adaptation

Statement dc:rights

• (c) 2011 Boon Pang Lim

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/24283

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/24283