engleski

Signal Denoising Using Wavelet Transformation with Variable Tresholding via Bayes Prediction and Ephraim-Malah Estimation

This paper introduces a new audio cleaning method which is composed of combination of stochastically and wavelet transformation based approach. This method could be implemented on signals which have been inherently contaminated with some degree of stationary noise. This paper focuses also on ability of discrete wavelet transformation (DWT) coefficients to have excellent resolution and transparency of distinguishing between noise and useful part of signal. Beside DWT, this paper focuses on stochastically approach which we need to ensure the information about minimum mean-square error (MMSE) of the spectral amplitude estimator. This kind of estimation will be performed on a silence- or pause- interval via Bayes prediction method and Ephraim-Malah estimation. This procedure provide us the degree of thresholding the DWT coefficients. The introduced approach results in a significant reduction of noise and will be compared with other noise-cleaning methods. All audio results and experimentation presented in this paper are performed using audio signals sampled at the professional sampling rates of 44.1 kHz or 48 kHz and quantized to 16 bits.

noise disturbances; audio material; Discrete Wavelet Transform; Bayes prediction; Eprhraim-Malah estimation

nije evidentirano