一种改进的谱减法语音增强算法

An Improved Spectral Subtraction Speech Enhancement Algorithm

传统的谱减法语音增强算法存在“音乐噪声”问题。针对此问题,引入过减因子和补偿因子两个参数对传统的谱减法进行改进。其中:过减因子可以调节在带噪语音功率谱中减去的噪声分量;补偿因子可以降低人耳对残留噪声的感知度。在两个参数的共同作用下,算法在抑制“音乐噪声”的同时可以使语音成分更加突出,且人耳的接受度更高。仿真结果表明,与传统的谱减法相比,使用改进形式的谱减法对带噪语音进行增强处理后,语音信号波形失真程度更低,信噪比改善更明显,MOS评分更高。

The traditional speech enhancement algorithm has the problem of“music noise”.To solve this problem,the traditional spectral subtraction method is improved by introducing two parameters:over subtraction factor and compensation factor.The over subtraction factor can adjust the noise component subtracted from the noisy speech power spectrum,and the compensation factor can reduce the human ear's perception of the residual noise.Under the combined action of the two parameters,the speech component can be more prominent while suppressing the“music noise”and the acceptability of human ears is higher.Simulation results indicate that,compared with the traditional spectral subtraction method,after using the improved form of spectral subtraction method to enhance the noisy speech,the distortion of the speech signal waveform is lower,the signal-to-noise ratio is improved more obviously,and the MOS score is higher.