Speech enhancement is becoming an important branch of speech signal process, which mainly used in noise reduction, preprocess of speech recognition and speech coding system. Various algorithms are supposed to improve the quality and intelligibility of speech. However, the two performance measures are sometimes conflicting. Therefore, a good speech enhancement system is a compromise between the two measures according to the requirement.This thesis firstly introduces the characteristics of speech, noise and human hearing. It discusses various of speech enhancement methods, then focuses on an improved spectral subtraction method based on masking effect.In order to reduce the musical residual noise and the background noise, a speech enhancement method based on masking properties of the human auditory system is described. This method uses BARK wavelet analysis to simulate the frequency feature of human auditory model to get the threshold. Experimental results have shown that this method performs better than conventional methods.In addition, according to the theory of wavelet transform and the idea of spectral subtraction as well as the hearing masking model, a modified spectral subtraction method on wavelet domain is presented in this thesis. It accords with human hearing characteristics while using BARK wavelet transform to analyze and synthesis speech signals. |