Abstract: An improved speech enhancement algorithm is proposed，which is based on a priori SNR estimation Wiener filtering.Firstly，the initial noise power spectrum NPS is estimated by computing the statistical average of the silent segments.Next，the power spectrum of the speech segments is estimated and the noise power spectrum is updated by smoothing the initial noise power spectrum and the noisy speech power spectrum NSPS.Lastly，the abrupt increases in noise level are taken into account and the power spectrum of the noise is adjusted adaptively according to the rate of the NSPS to NPS.The verification experiment between the proposed algorithm and other speech enhancement algorithm which based on short-time spectral estimation is implemented.The experiment result shows that the residual noise is reduced significantly and the speech distortion，and then the speech intelligibility are increased.

Key words: speech enhancement, Wiener filtering, a priori Signal to Noise Ratio（SNR）, decision directed

摘要： 提出了一种改进的语音增强算法，该算法以基于先验信噪比估计的维纳滤波法为基础。首先通过计算无声段的统计平均得到初始噪声功率谱；其次，计算语音段间带噪语音功率谱，并平滑处理初始噪声功率谱和带噪语音功率谱，更新了噪声功率谱；最后，考虑了某频率点处噪声急剧增大的情况，通过计算带噪语音功率谱与噪声功率谱的比值，自适应地调整噪声功率谱。将该算法与其他基于短时谱估计的语音增强算法进行了对比实验，实验结果表明：该算法能有效地减少残留噪声和语音畸变，提高语音可懂度。

关键词: 语音增强, 维纳滤波, 先验信噪比, 直接判决