Study of Speaker-Independent Mandarin Speech Recognition―Acoustic Phonetic Approach

摘要

This paper uses the acoustic-phonetic approach to develop a speaker-independent Mandarin word recognition system. Our main research topics include: the automatic segmentation of vowel and consonant, speaker adaptation techniques to overcome inter-speaker variations, and the recognition methods for vowels and consonants. The database for speakers are collected and classified into some speaker clusters. We use a continuous density HMM and modified Viterbi algorithm to characterize each PLU unit, it is simple and straightforward to add new features to the feature vector. In particular we focus on the techniques used to provide the acoustic-phonetic models of the subword units, and discuss the resulting system performance as a function of the type of acoustic modeling used.