PERFORMANCE EVALUATION OF DIFFERENT ASR CLASSIFIERS ON MOBILE DEVICE

Abstract

Automatic speech recognition is an option in contrast to composing on cell phones. Recently, it is usual and increasingly popular trend in communication. Classifier is used to classify the fragmented phonemes or words after the fragmentation of the speech signal. Several techniques are used for the classification of phoneme or word such as Neural Network, Support Vector Machine, Hidden Markov Model and Gaussian Mixture Model (GMM). This paper presents detailed study and performance analysis of above classification techniques. The performance evaluation is done to prove that GMM is better at the classification of signal data, and can be effectively used for improving the classification accuracy of the existing system. Our results show that accuracy of GMM is more than 20% better than other three classifiers. The performance of ASR classifier is evaluated on android phones, and evaluated for normal conversations in Hindi language used in day to day human to machine communications, using high-quality recording equipment.