Text independent speaker recognition system using Back Propagation Network with wavelet features

Automated speaker recognition system is extremely important in areas such as Forensic and Defense. Performance of an automated speaker recognition system is dependent on feature extraction and classification. As speech is a non-stationary signal and the information is present in low frequencies, it necessitates a non-stationary tool that performs multi resolution analysis. An exhaustive approach is carried out in this work to identify the wavelet that is best suited for feature extraction. Of the various wavelets, Discrete Meyer Wavelet provides higher inter-class variance and lesser intra- class variance. Sixteen features are extracted for wavelet co-efficients and a five layered Back Propagation Network is used for recognizing the speakers.