基于压缩感知的自适应谱减法语音增强算法

Adaptive spectral subtraction speech enhancement algorithm based on compressed sensing

针对在压缩感知框架下,噪声的影响会被扩大这个问题,提出了一种新的基于压缩感知的语音增强算法。该方案利用压缩感知下的行阶梯观测矩阵能够保留大部分语音特性的特点,对观测序列进行谱减法消噪,再对得到的观测序列进行基于输入信噪比的自适应重构,最后通过低通滤波器对重构语音进行平滑滤波,除去高频成分。实验结果表明:提出的语音增强方法具有较强的抗噪能力,重构速度快,输出的信噪比高,鲁棒性能好。

A novel speech enhancement algorithm based on compressed sensing is proposed for solving the problem that noise will be expanded under CS framework. Since row echelon measurement matrix can re- tain a large part of speech characteristics, the traditional spectral subtraction can be used to denoise the measurement sequence under the framework of compressed sensing. The reconstruction algorithm is modi- fied for different input signal-to-noise ratios （SNRs）. Finally, a low-pass filter is added to remove high frequency components. Simulation results indicate that the proposed algorithm has a high speech enhance- ment capability and can speed up the reconstruction noise intensities. Also, it performs high robustness under different