Speech Denoising Based On Perceptual Weighting Filter

The quality of speech occurs to degrade under the presence of a great variety of additive background noise. To improve the quality of speech communication and reduce the influence of the noise for the performance of speech communication, the speech denoising technique has been a hot subject in speech signal processing.In this paper, a new speech denoising algorithm for different noise environments based on perceptual weighting filter is proposed, which combines the spectral subtraction and exploits auditory perception properties in Wiener filter. The main contribution of the new algorithm is to modify the traditional Wiener filter by adopting the perceptual weighting function via the linear prediction technique. The method can redistribute the noise spectrum according to the clear speech. So the influence of the noise is further reduced and the quality is improved.By studying lots of references regarding speech enhancement technique systematically, the development and the current researches of speech enhancement are summarized in advance. And then several typical methods of speech enhancement are described in brief. The research idea is motivated according to the development of speech enhancement.The main research works of the paper are as follows:1. The principles of a few methods based on short-time spectrum estimation for speech enhancement are discussed. The performance of these methods are analyzed and compared by computer simulations.2. A new speech denoising algorithm based on perceptual weighting filter is proposed for white noise environment. The algorithm combines the spectrum subtraction and exploits auditory perception properties in Weiner filter. The simulation result indicates that the method is valid and feasible. 3. As a key point in speech enhancement algorithms, the noise estimation is discussed in this paper. In view of the shortcoming of the noise-estimation based on VAD (Voice Activity Detector), we focus on the fast noise-estimation algorithm and give the corresponding computer simulation.4. With the fast noise-estimation algorithm proposed above, we apply the speech denoising algorithm in the non-stationary environment. A new algorithm of speech denosing based on noise reshaping is proposed. we introduce a frequency compensation factor to compensate the uniformly effect of non-stationary background noise to the speech spectrum in the algorithm. The simulation result indicates that the approach can reduce the background noise effectively and improve the quality of speech.