JOURNAL OF INDIAN RESEARCH
VOLUME : 2, ISSUE : 4, OctoberDecember, 2014 (ISSN No. : 23214155) 

DESIGN & DEVELOPMENT OF DISCRETE HMM (DHMM) ISOLATED HINDI SPEECH RECOGNIZER 
Satish Kumar & Prof. Jai Prakash 

ABSTRACT 
This paper describes the insight of the design & development of a Proposed Hindi Speech Recognizer based on the discrete hidden Markov model (DHMM). Here we have proposed a new Quantizer which has been used with discrete hidden Markov modeling to get a proposed DHMM Hindi Speech Recognizer. The multidimensional Mel frequency cepstral coefficients(MFCC) speech vectors are converted into discrete symbols through a vector quantizer having size of the codebook as 32 i.e. each observation sequence for every utterance is represented by 32 code vectors or discrete symbols. The proposed design of a vector quantizer goes through various steps such as sorting, partitioning, quantizing and cluster indexing etc. to get an observation sequence. Each observation sequence is segmented into six states & a correct state sequence is found out by Viterbi algorithm. The parameter discrete symbol probability distribution (Emission matrix) and how a discrete symbol moves from one state to another (Transition matrix) are estimated by Baum Welch algorithm which makes use of forward & backward variables. The negative logarithm of probability distribution of discrete symbols in different states is used to find the index of the maximum probability which gives the recognized utterance. The results of the experimentation have shown that Proposed DHMM Speech Recognizer is more powerful and efficient as compared to other Speech Recognizer 

KEYWORD 
BaumWelch, DHMM, Markov model, Speech Recognizer, Viterbi, VQ. 

