A MFCC based Hindi Speech Recognition Technique using HTK Toolkit

To utilize the robot's capabilities, it is necessary for us to communicate with them efficiently. Thus, Human Robot Interaction is attracting the attention of most of the researchers these days. In this paper a speech recognition system has been developed using different feature extraction techniques like MFCC (mel frequency cepestral coefficient), LPC (linear predictive coding) and HMM (hidden markov model) is used as the classifier. Less work has been done for Hindi language in this field with a vocabulary size not very large. So, work in this paper has been done for Hindi database, with a vocabulary size a bit extended. HMM has been implemented using HTK Toolkit. Afterwards the performances of both of the techniques used have been compared. The work has been done using audacity for sound recordings and Cygwin to execute the HTK commands in Linux type environment in windows platform. As well as, the system developed has been tested in the speaker dependent and speaker independent both types of environments, whose performance results, as well as, the comparison graph of the system shows that MFCC performs well as compared to LPC in each and every condition.