基于无语音概率改进的对数谱幅度估计增强算法

针对谱减语音增强法中一直存在的去噪度、残留的音乐噪声和语音畸变度三者间均衡这一关键问题,本文提出一种基于无语音概率改进的对数谱估计增强算法.该算法结合无语音概率的思想,按照纯噪声帧和带噪语音帧两种状态.有区别地实时更新语音最小均方误差的对数谱增益,并利用无语音概率参数(SAP)自适应地调节平滑系数,以求随着噪声环境的变化,在去噪度、残留"音乐噪声"和语音畸变度之间自适应地折中.实验表明,该算法在相同去噪程度下,语音畸变和音乐噪声相对其他谱减法都同时地减弱,特别在低信噪比环境下优势更明显,而且平滑参数利用SAP参数,无需多余计算,便于实时处理.

An Improved Log-Spectral Amplitude Estimator Enhancement Method Based on Speech Absence Probability

This paper focuses on the trade-off among the amount of noise reduction,the level of musical residual noise and the speech distortion,which is the key problem of spectral subtraction speech enhancement. According to two different circumstances,pure noise frame and noisy speech frame,the novel algorithm modify the Log-Spectral Amplitude estimation separately. Then the smoothing- parameters in noise estimation are adaptively adjusted to the Speech Absence Probability (SAP) parameter which reflects stationary de- gree of the measured signal. The change of smoothing-parameters allows for automatic adaptation with various noisy environments and ob- tains the best trade-off. Experimental results demonstrate that the proposed algorithm has better performance of speech articulation and less musical noise. Moreover,this superiority is more significant at low SNR. Smoothing-parameters are dependent on SAP parameter,so that new algorithm needn't more redundant calculation than minimum mean-square error the Log-Spectral Amplitude estimation. It's good for real realization.