Wavelet based Human Voice Identification System

This paper investigates the use of wavelet transform in order to remove noise from the signals. One of the ongoing research in multimedia applications is speech signal processing. Wavelet denoising technique is attempted to reduce and remove noise from the audio signal. It is therefore required to transform audio signal to wavelet domain by using discrete wavelet transform followed by denoising algorithm. Both soft and hard thresholding of denoising technique is used to compare the performance of human noise identification. Denoising the signal is performed in the transformation domain and improvement in the denoising will be achieved in various families of wavelet transform. There are several types of wavelets such as Haar, Symlets, BiorSplines, Discrete Meyer (Dmey) and Reverse Biothogonal. In this paper, identification of human noise is verified with Dmey and Fejer-Korovkin wavelets. Comparative analysis is done to calculate SNR after the spectral subtraction. The quality of the audio signal is determined by Mean Square Error (MSE) and Signal to Noise Ratio (SNR) of the denoised signal. The simulation results are performed in Matlab program. From the result analysis, it is shown that the Fejer-Korovkin wavelet filter has a better performance as compared to other type of wavelet transform and can be able to identify and differentiate human voices with others.