基于听觉特性的迭代维纳滤波在语音分离中的应用

Iterative Wiener filtering based on the auditory characteristics for speech separation

本文讨论了引入人耳听觉特性的迭代维纳滤波在语音分离中的应用，即用矢量量化形成的码本反映目标话者的语音特征，通过计算滤波结果与这一特征的匹配度来模拟人耳在“鸡尾酒会效应”中的注意力机制．实验结果表明这一方法有很好的效果．

This paper discusses the application of Iterative Wiener Filtering(IWF) with Auditory Characteristics in Speech Separation. A codebook fOrmed by vector quantiza tion is used to represent the speech characteristics of the desired speaker, and human's attention in 'Cocktail Party Effect' is simulated by calculation of the matching degree between the filtering result and the codebook. Satisfactory speech separation results are obtained.