Effect of voice features cancellation in speaker identification system

Abstract

This paper introduces a good technique for the utilization of cancelable features for speaker identification. This technique depends on the use of encrypted features extracted from speech signals for identification. The encrypted features are obtained through convolution with random kernels. The proposed speaker identification system using cancelable features has been evaluated under the effect of Additive White Gaussian Noise (AWGN). Some quality metrics such as Log-Likelihood Ratio (LLR), Spectral Distortion (SD), and auto correlation have been used to assess the performance of the speaker identification system.