基于稀疏性的相位谱补偿语音增强算法

相位谱补偿语音增强算法通过调整相位谱对噪声进行压缩，提高重构信号的质量。针对传统的相位谱补偿(phase spectrum compensation, PSC)语音增强算法采用固定的相位补偿因子，且算法的性能易受噪声估计准确性的影响，提出了一种基于稀疏性的相位谱补偿(sparsity-based phase spectrum compensation, SPSC)语音增强算法。首先，利用噪声估计算法得到噪声幅度谱，利用基于幅度谱的语音增强算法得到目标语音幅度谱；接着，通过噪声和目标语音幅度谱之间的局部信噪比(Signal-to-Noise Ratio, SNR)来估计谱时间稀疏性；然后，利用sigmoid函数改进相位补偿因子，联合补偿因子和谱时间稀疏性，得到SPSC函数。最后，使用SPSC函数对相位谱中的谱分量进行补偿，通过短时傅里叶逆变换得到最终增强后的语音信号。仿真实验表明，在四种不同背景噪声的低信噪比下，新的相位谱补偿算法使增强语音获得了更好的LSD、PESQ和segSNR指标，说明新的算法在低信噪比下，可以有效恢复带噪语音中的语音成分，对噪声抑制效果明显，增强语音的质量和听感均有一定提升。 

Sparsity-based phase spectrum compensation for speech enhancement algorithm

The phase spectrum compensation speech enhancement algorithm compresses the noise by adjusting the phase spectrum to improve the quality of the reconstructed signal. For the traditional phase spectrum compensation (PSC) speech enhancement algorithm, a fixed phase compensation factor is adopted and the performance of the algorithm is easily affected by the accuracy of the noise estimation. A sparsity-based phase spectrum compensation (SPSC) for speech enhancement algorithm is proposed. Firstly, the magnitude spectrum of noise is obtained by using the noise estimation algorithm, the speech enhancement algorithm based on magnitude spectrum is used to obtain the magnitude spectrum of the target speech. Then, the spectro-temporal sparsity is estimated by the local signal-to-noise Ratio (SNR), which is obtained by the magnitude spectrum of the noise and target speech. Then, the sigmoid function is used to improve the phase compensation factor, and the SPSC function is obtained based on the compensation factor combined with the spectro-temporal sparsity. Finally, the SPSC function is used to compensate the spectral components in the phase spectrum, and the speech signal is finally enhanced by the inverse short-time Fourier transform. Simulation experimental results show that under the four different background noise with low SNR, the new phase spectrum compensation algorithm obtains better LSD, PESQ and segSNR indices, it shows that the new algorithm can effectively restore the speech components in the noisy speech under the low SNR, which has a significant effect on noise suppression, and improves the quality of speech and the audibility to some extent.