Chinese Vowels Recognition Method Based on Non-Homogeneous Hidden Markov Model

CHEN Li-wei,ZHAO Chun-hui,BAI Yu,SUN Yan
DOI: https://doi.org/10.3969/j.issn.1006-7043.2006.02.029
2006-01-01
Abstract:Introduced is a new speech recognition method,which is aimed at recognizing Chinese vowels and based on a non-homogeneous hidden Markov model(HMM). First extracted were the frequency responses of the vocal tract as the characteristic parameter,and then a non-homogeneous HMM was established to describe the real characteristics of speech more reasonably.The model resulting from this method was used in speech recognition experiments and compared with a homogeneous HMM.Results show this method of vowel recognition rate in speaker dependent mode can reach 98.73%,a distinct improvement over homogenous HMMs.This method provides good theoretical background and practical application value.
What problem does this paper attempt to address?