Design, Implementation and Evaluation of Hindi Speech Recognition system in Clean Environment

Abstract

The accuracy of speech recognition system depends on various external and internal factors. The internal factors mainly depend on front end modelling of speech recognition system like use of different feature extraction technique and algorithms. The external factors involve with parameters like environment (clean and noisy) and type of recognition; speaker dependent, speaker independent or speaker adaptation etc. Inside, a speech recognizer the rate of recognition finds a match between training and testing datasets. This paper deals with the two novel approaches, Voice Activity and Detection (VAD) and Differential Evolution (DE) to find the accuracy of speech recognition system in clean environment. Further, their performance is compared using Mel frequency Cepstral Coefficient (MFCC), Perceptual Linear Perceptron (PLP) and their hybridizations. The results show that voice activation and detection perform better than differential evolution under all feature extraction schemes.