A Noise Robust Speech Recognition System Using Wavelet Front End and Support Vector Machines

Recent works in speech recognition technology, classification techniques is focused on models, such as support vector machines (SVMs), in order to improve the generalization ability of the machine learning for noisy environments. However kernel function plays a vital role in the generalization ability of the SVMs. This paper address, the issue of noise robustness for an Automatic Speech Recognition (ASR) system focusing on a wavelet domain front end and SVM based classifier with different kernel functions. The proposed ASR has a front end that exploits the benefits of wavelet techniques for speech enhancement and feature extraction along with a comparison of different kernel functions for classification. The experiments are performed on speaker independent TIMIT database which are trained in a clean environment and later tested in the presence of AWGN for various Signal to Noise Ratio (SNR) levels. Experiments indicate that for large vocabulary the wavelet front end and the Radial Basis Function (RBF) kernel has more convergence area when compared to the polynomial kernel and the linear kernel for classification and robustness.