噪声环境下语音增强的算法分析与研究

Analysis and Study of Speech Enhancement Algorithm in Noise Environment

文章首先在谱减法的基础上提出了一种改进的加权幅度谱估计多带谱减法,改进的算法能够更好地抑制"音乐噪声",减少了语音谱的波动并提高了语音质量;其次从噪声类型、信噪比大小等方面分析比较改进的加权幅度谱估计多带谱减法(Mband)、最小均方误差对数幅度谱估计(MMSELSA)、维纳滤波法(Wiener)和最小值控制的递归平均算法(MCRA)4种语音增强效果。实验结果表明:在处理类语音噪声且在低信噪比环境下,有效性由高到低依次为:Mband、MMSE-LSA、Wiener、MCRA;在高信噪比条件下,MMSE-LSA增强效果较好。在处理低频带噪声时,有效性由高到低依次为:MMSE-LSA、Wiener、Mband、MCRA。MCRA对噪声的能量大小非常敏感,因此在处理非平稳噪声时增强效果相对较差。

An improved weighted amplitude spectrum estimation multi-band spectral subtraction based on spectral subtraction was proposed in this paper.Then,from the type of noise and signal to noise ratio to analyze the effective of four speech enhancement algorithm:an improved weighted amplitude spectrum estimation multiband spectral subtraction,Minimum Mean-Square Error Log-Spectral Amplitude Estimator,wiener filtering,Minima controlled recursive averaging.The experimental results show that:in the noise of phonological similarity and a low SNR environment,the effective from high to low is:an improved weighted amplitude spectrum estimation multi-band spectral subtraction,Minimum Mean-Square Error Log-Spectral Amplitude Estimator,wiener filtering,Minima controlled recursive averaging,under the condition of high SNR,Minimum MeanSquare Error Log-Spectral Amplitude Estimator enhancement effect is better;In dealing with the low frequency noise,the effective from high to low is:Minimum Mean-Square Error Log-Spectral Amplitude Estimator、wiener filtering、an improved weighted amplitude spectrum estimation multi-band spectral subtraction,Minima controlled recursive averaging.Minima controlled recursive averaging is sensitive to the energy of noise,so in dealing with the non-stationary noise,the enhancement effective is relatively poor.