Blind white denoising of speech signals

Discrete Wavelet Transform (DWT) has been used in the recent years in signal processing applications, i.e. filtering and compression. In the case of denoising because the energy of the noise is spread in the entire wavelet coefficients and it has low amplitude, it can be rejected by thresholding. In this paper, we propose a model to evaluate the influence of the denoising parameters in the quality of the speech signals, by a blind process. We examine the residual signal to establish an objective and blind criteria for selecting the following parameters: base, levels of decomposition, rule, and threshold. This model can be applied in any type of speech signal, no matter its behavior in time and frequency.