Title :
Channel and noise adaptation via HMM mixture mean transform and stochastic matching
Author :
Wong, Shuen Kong ; Shi, Bertram
Author_Institution :
Dept. of Electr. & Electron. Eng., Hong Kong Univ. of Sci. & Technol., Clearwater Bay, Hong Kong
Abstract :
We present a non-linear model transformation for adapting Gaussian mixture HMMs using both static and dynamic MFCC observation vectors to additive noise and constant system tilt. This transformation depends upon a few compensation coefficients which can be estimated from channel distorted speech via maximum-likelihood stochastic matching. Experimental results validate the effectiveness of the adaptation. We also provide an adaptation strategy which can result in improved performance at reduced computational cost compared with a straightforward implementation of stochastic matching
Keywords :
Gaussian processes; computational complexity; hidden Markov models; maximum likelihood estimation; speech recognition; transforms; Gaussian mixture HMMs; HMM mixture mean transform; adaptation strategy; additive noise; channel adaptation; channel distorted speech; compensation coefficients; computational cost; dynamic MFCC observation vectors; maximum-likelihood stochastic matching; noise adaptation; nonlinear model transformation; performance; static MFCC observation vectors; stochastic matching; Additive noise; Computational efficiency; Hidden Markov models; Maximum likelihood estimation; Mel frequency cepstral coefficient; Nonlinear distortion; Nonlinear dynamical systems; Speech; Stochastic processes; Stochastic resonance;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1999. Proceedings., 1999 IEEE International Conference on
Conference_Location :
Phoenix, AZ
Print_ISBN :
0-7803-5041-3
DOI :
10.1109/ICASSP.1999.758122