DESIGN & DEVELOPMENT OF CONTINUOUS DENSITY HMM (CDHMM) ISOLATED HINDI SPEECH RECOGNIZER

Abstract

This paper describes the insight of the design & development of a Proposed Hindi Speech Recognizer based
on the continuous density hidden Markov model (CDHMM) . Here we have proposed a new recognizer which have been used
with continuous density hidden Markov modeling to get a proposed CDHMM Hindi Speech Recognizer. The
multidimensional Mel frequency cepstral coefficients(MFCC) speech vectors are extracted from raw speech for every given
word that are used as a sequence of observation vectors, are uniformly segmented into 6 states . For each state Gaussian
Mixture Model(GMM) parameters such as a Mean(
 jk
) & Covariance (
 jk
) matrix & number of Mixtures(
Cjk
) are
calculated and simultaneously hidden Markov Model parameters such as
   ( , , ) A B
are calculated to prepare a GMMHMM Model known as cdhmm , { , , }    Cjk jk jk  . Here ,Q=6, Mixture Components(K or M)=16, The covariance
matrix used can be full or diagonal type matrix. Investigaions are done in this paper to find the optimal number of Gaussian
mixture components that gives maximum accuracy in the context of Hindi speech recognition system. The results of the
experimentation have shown that Proposed CDHMM Speech Recognizer gives maximum performance when no. of Gaussian
Mixture Model Components used is 16. This method is more powerful and efficient as compared to discrete Speech
Recognizer.