MULTIPLE VQ HIDDEN MARKOV MODELING FOR SPEECH RECOGNITION

In this paper a new variant of HMM, named Multiple VQ HMM (MVQHMM), is presented. Its main characteristic is the use of a separate codebook for each model. Procedures for training and probability evaluation of these models are described. The evaluation procedure combines the quan tization distortions of the vector sequences with the discrete HMM gen eration probabilities. Comparative results on an isolated word recogni tion system are shown, between MVQHMM and discrete and semi-continuous HMM. These results show that using separate codebooks and including t he quantization distortion in the decision criterion improve the perfo rmance of the system. Furthermore, the multiple VQ hidden Markov model s seem to be more robust than the discrete and semi-continuous ones in relation to the inter-speaker variability of the recognition system.