POSTPROCESSOR USING FUZZY VECTOR QUANTIZER IN HMM-BASED SPEECH RECOGNITION

Abstract

A postprocessor using fuzzy vector quantisation (FVQ) in HMM-based speech recognition is presented. A method to reduce the amount of computation in the FVQ postprocessor is also proposed. We show that the proposed method results in a higher recognition rate with almost the same amount of computation compared to the conventional Viterbi scoring method.