Abstract: The theory of wavelet denoising is analysised,and a wavelet speech enhancement method based on improved－threshold is proposed according to the characteristics of random noise wavelet transform on the different scale and the relationship between the noise Lipschitz exponents and its wavelet transform maximum modulus,also considering the unvoiced and voiced speech’s features．A parameter can be adjusted properly in the new threshold function to obtain the best estimations of the wavelet coefficients which enables this function to become a more general case incorporating the hard and soft threshold.Using improved－threshold to process the wavelet coefficients of the speech containing noise,not only the noise can be removed,but also the quality degradation of the unvoiced sounds can be prevented．Simulation results demonstrate the method is very effective in speech enhancement．

Key words: wavelet transform, speech signal, threshold, denoising

摘要： 分析了小波去噪原理,根据随机噪声的小波变换系数在不同尺度上的传递特性和噪声信号奇异性与小波模极大值的关系,同时考虑到语音中浊音和清音的特点,提出了一种改进阈值的小波域语音增强方法。在阈值函数中引入参数,通过调整参数以获得最佳的小波系数的阈值估计,使得改进阈值介于硬阈值与软阈值之间。利用改进阈值对染噪语音信号的小波系数进行阈值处理,既抑制了噪声,又减少了语音段信息的损失。仿真结果表明,这是一种有效的语音增强方法。

关键词: 小波变换, 语音信号, 阈值, 去噪