Different speech enhancement algorithms have their inherent respective advantages and disadvantages.The currently applied speech enhancement algorithms are lack of systematic performance evaluation, so finding a reasonable performance evaluation method for the speech enhancement algorithms is beneficial to improve the algorithms. Taking into all-round consideration the speech enhancement algorithms based on different principles and the environment of pink noises, the paper selects four classic and representative speech enhancement algorithms: the wiener filtering method, minimum mean square error method, signal subspace method and wavelet transform method. On this basis, the paper uses the combined segment SNR and perceptual quality evaluation method to evaluate the selected algorithms in respect of their performance indexes such as the noise reduction level ,distortion and intelligibility. Simulation results show that the new method can evaluate the various speech enhancement algorithms generally and rationally in consistent with the subjective feelings of human ears, so it is proved to be of good practicality.%各种不同的语音增强算法都有各自的优缺点,对于现有的语音增强算法的存在,缺少系统化的性能评价,找出合理评价语音增强算法性能的方法有利于对算法进行改进.本文全面考虑基于不同原理的语音增强算法,针对粉红噪声环境,选取各类算法中经典的具有代表性的4种语音增强算法:维纳滤波法、最小均方误差法、信号子空间法及小波变换法,在此基础上运用分段信噪比与语音感知质量评价算法相结合的方案来综合评价语音增强算法在噪声消除水平、失真程度和可懂度等语音质量方面的性能指标.仿真结果表明这种新方法能较合理地评价出各种语音增强算法的综合性能,与人耳的主观感受相一致,从而更具有实用性.
展开▼
