FLEXIBLE PARAMETER TYING FOR CONVERSATIONAL SPEECH RECOGNITION

Modeling pronunciation variation is key for recognizing conversational speech. Previous efforts on pronunciation modeling by modifying dictionaries only yielded marginal improvement. Due to complex interaction between dictionaries and acoustic models, we believe a pronunciation modeling scheme is plausible only when closely coupled with the underlying acoustic model. This paper explores the use of flexible parameter tying for pronunciation modeling. In particular, two new techniques are investigated: Gaussian tying and flexible tree clustering. We report a 1.3% absolute WER improvement over the traditional modeling framework on the Switchboard task.

[1]  Michael Finke,et al.  Wide context acoustic modeling in read vs. spontaneous speech , 1997, 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing.

[2]  William J. Byrne,et al.  Pronunciation modelling using a hand-labelled corpus for conversational speech recognition , 1998, Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181).

[3]  S. J. Young,et al.  Tree-based state tying for high accuracy acoustic modelling , 1994 .

[4]  Thomas Hain,et al.  IMPLICIT PRONUNCIATION MODELLING IN ASR , 2002 .

[5]  Xiuyang Yu,et al.  What kind of pronunciation variation is hard for triphones to model? , 2001, 2001 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.01CH37221).

[6]  Atsushi Nakamura.  Restructuring Gaussian mixture density functions in speaker-independent acoustic models , 2002, Speech Commun..

[7]  Harriet J. Nock,et al.  Pronunciation modeling by sharing gaussian densities across phonetic models , 1999, EUROSPEECH.