Noise robust speaker verification using paralel model combination and local features
Interfering noise severely degrades the performance of a speaker verification system. The Parallel Model Combination (PMC) technique is one of the most efficient techniques for dealing with such noise. Another method is to use features local in the frequency domain. Recently, we proposed Mel-Frequency Discrete Wavelet Coefficients (MFDWCs)  as speech features local in frequency domain. In this paper, we discuss using PMC along with MFDWC features to take advantage of both noise compensation and local features (MFDWCs) to decrease the effect of noise on verification performance. We evaluate the performance of MFDWCs for various noise types and noise levels. We also compare the performance of these versus MFCCs and both using PMC for dealing with additive noise. The experimental results show significant performance improvements for MFDWCs versus MFCCs after compensating the HMMs using the PMC technique. For example the MFDWCs gave 6.29 points performance improvement on average over MFCCs for 12 dB. This corresponds to 38.33% error reduction.