Recognition of phonetic Arabic figures via wavelet based Mel Frequency Cepstrum using HMMs

Abstract

This work describes the recognition of phonetic Arabic figures. Speech recognition technology has made steady progress in its 50years history and has succeeded in creating several substantial applications. The goal of speech recognition research is to produce a machine which will recognize accurately the normal human speech from any speaker.To improve the performance of recognition system, an effective and robust technique is proposed to extract speech feature. The input speech signal is decomposed into various frequency channels based on time-frequency multi-resolution property of wavelet analysis. For capturing the characteristics of the signal, the Mel-Frequency Cepstrum Coefficients “MFCCs” of the wavelet channels are calculated. Hidden Markov Models “HMMs” were used for the recognition stage. Different forms of wavelet functions were used to evaluate the best wavelet signal to extract the best features of the signals. It is found that the wavelet signal “db8” gives the highest values of recognition accuracy rate. A recognition rate of 98% was obtained using the proposed feature extraction technique.A comparison between different features of speech is given. The features based on the Cepstrum give accuracy of 94% for speech recognition while the features based on the short time energy in time domain give accuracy of 92%. The features based on formant frequencies give accuracy of 95.5%. It is clear that the features based on MFCCs with accuracy of 98% give the best accuracy rate. So the features depend on MFCCs with HMMs may be recommended for recognition of the spoken Arabic digits.