基于自适应MMSE-LSA与NMF的语音增强算法

Speech Enhancement Algorithm Based on Adaptive MMSE-LSA and NMF

针对目前已有的语音增强算法在低信噪比的非平稳噪声环境下存在语音增强性能欠佳、计算复杂度高、语音失真与音乐噪声的问题,提出基于自适应对数谱幅度的最小均方误差(MMSE-LSA)与非负矩阵分解(NMF)的语音增强算法。首先利用自适应MMSE-LSA估计器对含噪语音信号进行增强,以提高输入信号的信噪比,接着对增强后产生的语音失真和残留噪声利用NMF算法进行补偿,既保证语音质量,又尽可能地消除噪声干扰。仿真实验结果表明,语音增强算法与经典的谱减算法、维纳滤波算法相比,不仅提高了输出信噪比,而且降低了音乐噪声,在可懂度和清晰度方面均具有较明显的优势。

Aiming at the problems of poor speech enhancement performance,high computational complexity,speech distortion and music noise in the existing speech enhancement algorithms in a non-stationary noise environment with low SNR,a speech enhancement algorithm based on adaptive minimum mean square error of logarithmic spectral amplitude(MMSE-LSA)and non-negative matrix factorization(NMF)was proposed.Firstly,the adaptive MMSE-LSA estimator was used to enhance the noisy speech signal to improve the signal-to-noise ratio of the input signal.Then,the NMF algorithm was used to compensate the speech distortion and residual noise generated after the enhancement,which not only ensured the speech quality,but also eliminated the noise interference as much as possible.Simulation results showed that,compared with the classical spectral subtraction algorithm and Wiener filter algorithm,the proposed algorithm not only improved the output signal-to-noise ratio,but also reduced the music noise,and had obvious advantages in intelligibility and clarity.