Stationary wavelet Filtering Cepstral coefficients (SWFCC) for robust speaker identification

Abstract

Extracting robust effective speech features is one of the challenging topics in the speaker recognition field, especially in noisy conditions. It can substantially improve the robustness recognition accuracy of persons from their voice signals against such conditions. This paper proposes a new feature extraction approach called Stationary Wavelet Filtering Cepstral Coefficients (SWFCC) for noisy speaker recognition. The proposed approach incorporates a Stationary Wavelet Filterbank (SWF) and an Implicit Wiener Filtering (IWF) technique. The SWF is based on the stationary wavelet packet transform, which is a shift-invariant transform. The performance of the proposed SWFCC approach is evaluated on the TIMIT dataset in the presence of different types of environmental noise, which are taken from the Aurora dataset. Our experimental results using the Gaussian Mixture Model-Universal Background Model (GMM-UBM) as a classifier show that SWFCC outperforms various feature extraction techniques like MFCC, PNCC, and GFCC in terms of recognition accuracy.