Title :
A unified maximum likelihood approach to acoustic mismatch compensation: application to noisy Lombard speech recognition
Author :
Afify, Mohamed ; Gong, Yifan ; Haton, Jean-Paul
Author_Institution :
CRIN-INRIA Lorraine, Vandoeuvre-les-Nancy, France
Abstract :
In the context of continuous density hidden Markov model (CDHMM) we present a unified maximum likelihood (ML) approach to acoustic mismatch compensation. This is achieved by introducing additive Gaussian biases at the state level in both the mel cepstral and linear spectral domains. Flexible modelling of different mismatch effects can be obtained through appropriate bias tying. A maximum likelihood approach for joint estimation of both mel cepstral and linear spectral biases from the observed mismatched speech given only one set of clean speech models is presented, where the obtained bias estimates are used for the compensation of clean speech models during decoding. The proposed approach is applied to the recognition of noisy Lombard speech, and significant improvement in the word recognition rate is achieved
Keywords :
Gaussian processes; acoustic signal processing; cepstral analysis; decoding; hidden Markov models; maximum likelihood estimation; noise; spectral analysis; speech processing; speech recognition; CDHMM; acoustic mismatch compensation; additive Gaussian biases; bias estimates; clean speech models; continuous density hidden Markov model; decoding; joint estimation; linear spectral bias; linear spectral domain; mel cepstral bias; mel cepstral domain; mismatch effects modelling; noisy Lombard speech recognition; observed mismatched speech; state level; unified maximum likelihood approach; word recognition rate; Acoustic applications; Acoustic noise; Additive white noise; Cepstral analysis; Hidden Markov models; Maximum likelihood estimation; Parameter estimation; Speech enhancement; Speech recognition; Working environment noise;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1997. ICASSP-97., 1997 IEEE International Conference on
Conference_Location :
Munich
Print_ISBN :
0-8186-7919-0
DOI :
10.1109/ICASSP.1997.596064