基于多窗谱估计的改进维纳滤波语音增强

摘要

为减少维纳滤波在语音增强中残留的"音乐噪声",将多窗谱估计和改进的维纳滤波方法结合,并进行语音合成.设计了基于多窗谱估计的改进维纳滤波语音增强方法,该方法采用多窗谱估计噪声功率谱,改进维纳滤波降噪得到增强语音,以及重叠相加语音合成,并给出仿真对比验证.结果表明,基于多窗谱估计的改进维纳滤波方法在抑制噪声,减少音乐噪声方面优于基于维纳滤波的增强算法和基于多窗谱估计的改进谱减法的增强算法.%In order to reduce the musical residual noise left by the speech enhancement of wiener filter, multi-taper spectrum estimation is combined with the improved wiener filter method while processing speech synthesis.A wiener filter speech enhancement method based on multi-taper spectrum estimation is designed.The method adopts the noise power spectrum of multi-taper spectrum estimation, improves wiener filter to reduce noise and enhance speech, and overlap-adds it to achieve speech synthesis.Then the simulation contrast verification is given.The result shows that the wiener filter speech enhancement method based on multi-taper spectrum estimation performs better in noise suppression and musical noise reduction than speech enhancement algorithm based on wiener filter as well as improved spectral subtraction method based on multi-taper spectrum estimation.