The Use of Wavelets in Speaker Feature Tracking Identification System Using Neural Network

Continuous and Discrete Wavelet Transform (WT) are used to create text-dependent robust to noise speaker recognition system. In this paper we investigate the accuracy of identification the speaker identity in nonstationary signals. Three methods are used to extract the essential speaker features based on Continuous, Discrete Wavelet Transform and Power Spectrum Density (PSD). To have better identification rate, two types of Neural Networks (NNT) are studied: The first is Feed Forward Back Propagation Neural Network (FFBNN) and the second is perceptron. Up to 98.44% identification rate is achieved. The presented system depends on the multi-stage features extracting due to its better accuracy. The multistage features tracking based system shows good capability of features tracking for tested signals with SNR equals to -9 dB using Wavelet Transform, which is suitable for non-stationary signal. Key-words — Speaker identification; Continuous and discrete wavelet transform; Linear prediction coefficient; and text-dependent.