5th International Conference on Spoken Language Processing

Sydney, Australia
November 30 - December 4, 1998

A Nonstationary Autoregressive HMM With Gain Adaptation For Speech Recognition

Ki Yong Lee (1), Joohun Lee (2)

(1) Soongsil University, Korea
(2) Dong-Ah Broadcasting College, Korea

In this paper, a time domain approach for speech recognition is developed. The nonstationary autoregressive (AR) hidden markov model (HMM) with gain contour is proposed for modeling the statistical characteristics of the speech signal. The parameter of nonstationary AR model was modeled by the polynomial function with linear combination of M known basis functions. In this proposed model, speech signal is blocked by samples into fixed-length frames and modeled by nonstationary AR model controlled by markov switching sequences at each frame. Given the HMM parameter set of the speech, the gain-adapted recognition algorithm is developed for speech recognition.

Full Paper

Bibliographic reference.  Lee, Ki Yong / Lee, Joohun (1998): "A nonstationary autoregressive HMM with gain adaptation for speech recognition", In ICSLP-1998, paper 0408.