An efficient Automatic Speaker Identification System based Pitch Frequency Estimation in Degraded Environmental Conditions

Abstract

In this paper, we investigate the utilization of the Normalized Pitch Frequency (NPF) as an extracted feature from speech signals to be combined with the Mel Frequency Cepstral Coefficients (MFCCs) and polynomial coefficients. The objective is to compose more robust feature vectors to various forms of degradation such as Additive White Gaussian Noise (AWGN) and music interference. A matching process is performed to determine the identity of the unknown speaker, using a trained Artificial Neural Network (ANN) as a classifier. An Automatic Speaker Identification (ASI) system is presented in this paper comprising pre-processing methods based on Discrete Transforms (DTs) such as the Discrete Cosine Transform (DCT), the Discrete Sine Transform (DST), and the Discrete Wavelet Transform (DWT) for presenting robust features. Speech enhancement techniques such as Spectral Subtraction, Wiener filtering, adaptive Wiener filtering, and wavelet denoising are investigated to mitigate the impact of noise and improve the efficiency of the ASI system. Simulation results demonstrate that the utilization of the NPF with MFCCs as features extracted from both the speech signals and the DCTs of these signals increases the ASI system accuracy in the presence of noise and interference. The wavelet denoising enhances the proposed system effectiveness and gives high recognition rates even with very low Signal-to-Noise Ratios (SNRs).