非平稳环境下基于人耳听觉掩蔽特性的语音增强

Speech Enhancement based on Human Auditory Masking Properties under Non-stationary Environments

传统的语音增强算法往往仅对平稳噪声或缓慢变化的噪声有效,且残留的音乐噪声较大。对此,本文研究了一种非平稳环境下基于听觉掩蔽效应的语音增强算法。该算法对传统谱减法的功率谱估计算法进行改进,根据最小均方误差原则和语音信号的听觉掩蔽阈值调整功率谱估计的参数,并引入了基于最小值统计特性的噪声估计算法,使估计的噪声更好地跟踪噪声的变化。实验结果表明:该算法对平稳和非平稳的噪声都得到较好的增强效果,且较好地抑制了音乐噪声。

This paper addresses the problem of single channel speech enhancement under stationary and non-stationary environments, which based on the masking properties of human auditory system. This algorithm can overcome the deficiency of the conventional speech enhancement algorithms, which were only efficient for stationary environments and have large level of musical residual noise. During the estimation of power spectrum of the speech, the parameters of the estimator can be modified by the MMSE and the masking threshold of the speech, by this way, we can find the best trade off among the amount of noise reduction, the speech distortion and the level of musical residual noise. For the best tracking the variation of the environment, the method of minimum statistics was introduced for noise power spectrum estimation. Objective and subjective evaluation of the proposed algorithm is performed with several noise types in the Noisex-92 database with different time frequency distributions. The evaluations confirm that the enhanced speech by proposed algorithm is more pleasant to a human listener for every noise conditions.