EAGER: Variance and Invariance in Voice Quality

点击翻译按钮获取中文摘要

This EArly Grant for Exploratory Research aims at developing a database and model of voice source variability. A model of voice variations across people and speech tasks could improve the naturalness of speech synthesis systems. In addition, understanding what aspects of the voice, if any, are speaker-specific, should aid in developing better speaker identification and verification algorithms. Knowing how much a person could change his or her voice quality without compromising their vocal identity, could also inform medical rehab applications. A better understanding of the human voice will, thus, be of significant impact scientifically, and for engineering and medical applications. The project has strong outreach and dissemination programs and fosters interdisciplinary activities in Electrical Engineering, Linguistics, and Speech and Hearing Science at UCLA. It will train undergraduate and graduate students in important cross-disciplinary activities of technological and scientific significance. This exploratory project will analyze and discover how the voice source varies within and across talkers under circumstances that introduce variability in everyday life situations. The project aims to address three questions: 1) Does an individual talker's voice source vary significantly across recording sessions and speech tasks?, 2) Do bilingual talkers show more or less intra-talker variation when speaking in English?, and 3) Most importantly, how does intra-talker variability from all these sources compare with inter-talker variability? Understanding these issues will require a high-quality speech database with multiple voice samples from many talkers (in this case 200) which will be collected and distributed to other researchers. Acoustic analyses will reveal inter- and intra-talker variability in the voice source across different situations by generating a multi-dimensional acoustic profile of each talker that specifies the range of parameter values that are typical in the corpus for that talker, and the likelihood of deviations from that usual profile.