Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Prior Parameter Transformation for Unsupervised Speaker Adaptation

Guoqiang Li, Limin Du, Ziqiang Hou

Institute of Acoustics, Chinese Academy of Sciences, Beijing, China

In a strictly Bayesian approach, prior parameters are assumed known, based on common or subjective knowledge. But a practical solution for maximum a posteriori adaptation methods is to adopt an empirical Bayesian approach, where the prior parameters are estimated directly from training speech data itself. So there is a problem of mismatches between training and testing conditions in the use of prior parameters. We proposed a prior parameter transformation (PPT) adaptation approach that transforms the prior parameters to be more representative of the new speaker. In this paper we extend it to unsupervised mode. For easily confused speech units, different transformation matrices are applied to make them distinct. Initial experiments show that the PPT algorithm can get much improvement for a small amount of adaptation data even in the unsupervised mode.

Full Paper

Bibliographic reference.  Li, Guoqiang / Du, Limin / Hou, Ziqiang (2000): "Prior parameter transformation for unsupervised speaker adaptation", In ICSLP-2000, vol.3, 698-701.