语音信号的频域处理,语音虽然是一个时变、非平稳的随机过程。但在短时间内可近似看作是平稳的。因此如果能从带噪语音的短时谱中估计出“纯净”语音的短时谱，即可达到语音增强的目的。由于噪声也是随机过程，因此这种估计只能建立在统计模型基础上。利用人耳感知对语音频谱分量的相位不敏感的特性，这类语音增强算法主要针对短时谱的幅度估计。

-voice signals in the frequency domain processing, voice is a time-varying, nonstationary random process. But in a short period of time can be approximated as smooth. So if Noisy Speech from the short-term spectrum estimate "pure" voice of the short-term spectrum, and reached speech enhancement purposes. As the noise is random process, which can only be estimated based on statistical models based on. Use ear perception of voice spectrum component of the phase sensitive to the characteristics of such speech enhancement algorithms targeted at the rate of short-term spectral estimation.