Abstract: Hindi is a language of masses. It is the national language of India and widely popular in Indian sub-continent. It is important to develop interactive system which comprehend Hindi language. This article presents the implementation of Hindi speech recognition system. The systems have been developed on HTK-Toolkit V 3.4 in Linux environment (Ubuntu 10.04.3 LTS). The Hindi speech recognition system has been developed for 101-word vocabulary size. Each word is uttered for a number of times to capture all the acoustic variability’s. The system has been developed in two parts namely front-end and back-end. Front-end part covers preprocessing and feature extraction while back-end covers acoustic modeling, language modeling and recognition.The comparative analysis shows that MFCC perform better in same training and testing conditions while PLP perform better in mismatch conditions while both the feature extraction techniques outperform LPCC.
Keywords: MFCC, LPCC, PLP, HTK, Hindi Speech Recognition Engine.