PERFORMANCE ENHANCEMENT OF DCT BASED SPEAKER RECOGNITION USING WAVELET DE-NOISING

Munmuni Patir, V.K Govindan

Abstract: Presence of noise in the speech signal is one of the major problems in Speaker Recognition. The speaker recognition performance gradually degrades as the intensity of noise increases. The system gives high accuracy when the speech signal is noise free, but in real life scenario getting a noise free speech signal is challenging. Hence, elimination of the noise from speech signal is an important aspect in speaker recognition process. This work uses wavelet based denoising of the recorded speech signal in order to enhance the performance of speaker recognition. In this paper, wavelet based denoising technique has been applied to the DCT based speaker recognition system which was proposed in our previous work. Additive white Gaussian noise has been added to the speech signal and performance analysis of the system has been done using different SNR value

Keywords: Wavelet denoising, AWGN, Speaker Recognition, Thresholding, DCT, Feature Extraction.

DOI: https://doi.org/10.15623/ijret.2015.0404129