基于改进阈值函数的小波包语音增强算法的研究

基于小波分析的阈值语音信号去噪技术是语音信号处理的一个重要热点技术，由于语音信号的非平稳性，传统的小波阈值去噪算法虽然能够衰减一部分语音信号中的噪声，但这些算法会不可避免地造成有用语音信号的损失。小波包可以同时对语音的低频和高频部分进行分解，能更好保留语音信息，减少噪声对语音的影响；同时，文中提出一种新的阈值函数，它对噪声清除得更加干净。仿真结果显示，该阈值函数较软、硬阈值函数具有明显的优越性。

Speech Enhancement Algorithm Based on a Modified Threshold Function

Threshold de-noising technology for speech signal based wavelet analysis is an important topic in speech signal processing. Due to the nonstationarity of speech signals, traditional Wavelet Threshold De-noising Algorithm could attenuate part of the noise in speech signal. However, this algorithm would inevitably cause the loss of the useful speech signal. By using wavelet packet the speech signal could be decomposed into the low frequency part and the high frequency part, thus the speech information could be better, kept while the influence of noise be eliminated. And a new threshold function is proposed, which could de-noise the speech signal more completely. Simulation results show that, this proposed function, as compared to the hard or soft threshold function is of obvious superiority.