A comparative study of parametric coding and wavelet coding based feature extraction techniques in recognizing spoken words

Abstract

Speech recognition is a fascinating application of digital signal processing offering unparalleled opportunities. In this paper, a comparative study of different feature extraction techniques like Linear Predictive Coding (LPC), Discrete Wavelet Transforms (DWT) and Wavelet packet Decomposition (WPD) are employed for recognizing speaker independent spoken isolated words. Voice signals are sampled directly from the microphone and then they are processed using these three techniques for extracting the features. Words from Malayalam, one of the four major Dravidian languages of southern India are chosen for recognition. Training, testing and pattern recognition are performed using Artificial Neural Networks (ANN). This work includes three speech recognition methods. First one is a hybrid approach with LPC and ANN, second method uses a combination of DWT and ANN and the third one utilizes a combination of WPD and ANN. Back propagation method is used to train the ANN. The proposed method is implemented for 50 speakers uttering 20 isolated words each. All the three methods produce good recognition accuracy. LPC based method produced an accuracy of 81.20%, DWT gave an accuracy of 90% and WPD produced a recognition accuracy of 87.50%. Thus wavelet based methods are found to be more suitable for recognizing speech because of their multi-resolution characteristics and efficient time frequency localizations. Moreover, wavelet methods have a better capability to model the unvoiced sound details.