USE OF MULTIPLE VECTOR QUANTIZATION FOR SEMICONTINUOUS-HMM SPEECH RECOGNITION

Although the continuous hidden Markov model (CHMM) technique seems to be the most flexible and complete tool for speech modelling, it is not always used for the implementation of speech recognition systems beca use of several problems related to training and computational complexi ty. Thus, other simpler types of HMMs, such as discrete (DHMM) or semi continuous (SCHMM) models, are commonly utilised with very acceptable results. Also, the superiority of continuous models over these types o f HMMs is not clear. The authors' group has recently introduced the mu ltiple vector quantisation (MVQ) technique, the main feature of which is the use of one separated VQ codebook for each recognition unit. The MVQ technique applied to DHMM models generates a new HMM modelling (b asic MVQ models) that allows incorporation into the recognition dynami cs of the input sequence information wasted by the discrete models in the VQ process. The authors propose a new variant of HMM models that a rises from the idea of applying MVQ to SCHMM models. These are SCMVQ-H MM (semicontinuous multiple vector quantisation HMM) models that use o ne VQ codebook per recognition unit and several quantisation candidate s for each input vector. It is shown that SCMVQ modelling is formally the closest one to CHMM, although requiring even less computation than SCHMMs. After studying several implementation issues of the MVQ techn ique, such as which type of probability density function should be use d, the authors show the superiority of SCMVQ models over other types o f HMM models such as DHMMs, SCHMMs or the basic MVQs.