一种单通道语音倒谱域去混响新方法

大多数去混响的算法都是基于短时分析,不可避免地带来了截断效应;如果对混响语音进行逆滤波,则要求房间冲激响应满足最小相位.本文提出了一种针对单通道的语音长时分帧的去混响算法.算法采用长时帧分析,有效减小了截断效应的影响.首先对混响语音信号进行长时分帧计算其最小相位分量倒谱,然后根据纯净语音和房间冲激响应倒谱的不同分布,在倒谱域阻带滤波得到去混响语音最小相位分量的估计,根据最小相位倒谱理论,进而可以估计出去混响语音频域的模,最后利用人耳对相位信息的不敏感性,用估计的去混响语音的模和混响语音的相位合成,傅里叶反变换得到去混响语音时域波形.实验表明,该算法有良好的去混响效果.

A new method of speech dereverberation of single channel with cepstral processing

Most of the algorithms of dereverberation are based on short term analysis,which inevitably bring the truncation error. If we dereverberate the reverberation voice by inverse filter, room impulse response is limited to minimum phase. A noval single channel algorithm of dereverberation using long term frame analysis is proposed in this paper. First, the minimum phase cepstrum component of the reverberation speech signal was calculated in long term frame. Then the stop band filter was used to get the minimum phase component of dereverberation speech in the cepstral domain. Afterwards, mod of dereverberation speech in frequency domain can be estimated according to minimum phase cepstrum theory. Finally, since the human ear is insensitive to the phase information, the dereverberation speech can be synthesized with the estimated mod and the phase of reverb speech signal in frequency domain. Then the time domain waveform of dereverberated voice was obetained by inverse Fourier transform. Experiments show that the algorithm has a good dereverberated performance.