De-noising of Audio Signal using Heavy Tailed Distribution and comparison of wavelets and thresholding techniques

This paper presents a simple and novel approach for de-noising of the Audio Signals i.e. non-stationary signal using statistical distribution function at different sub-band level of coefficients. The performance of wavelets are analysed under various thresholding techniques. Nonstationary signals are continuous in nature consequently we use 1D Discrete Wavelet Transform which gives us a better time- frequency localization as compared to the spectral analysis of Fourier Transform. The coefficients of wavelet are modelled on the basis of Heavy Tailed Distribution function which gives a valuable and stable representation against Gaussian distribution function in filtering noise components from the signal. We have used the statistically independent White Noise of magnitude 5db to make the noisy audio signal. The performance is numerically assessed in terms of Signal-to-Noise ratio (SNR) and Mean-Square error (MSE) terms. The Coiflets wavelet in combination with Neighbouring Coefficients with Level-Dependent Threshold Estimator shows superior performance in comparison to other wavelets.