基于计算听觉场景分析的混合语音分离

Mixture Speech Separation Based on Computational Auditory Scene Analysis

基于计算听觉场景分析的理论,使用onset/offset线索完成混合语音分离研究。将经过外围模型处理后的数据,在时域和频域上分别检测并匹配onset/offset,利用时频图上的onset/offset信息合并片段,实现语音分离。通过对3类混合语料进行实验分析,得出onset/offset线索可以同时处理清音和浊音,对声音混合类型没有限制,能得到较好的分离效果。

Based on the principle of Computational Auditory Scene Analysis（CASA）,this paper describes a model of speech separation by analyzing onset and offset of auditory events.The model detects onsets and offsets,and generates segments by matching corresponding onset and offset fronts.The model is evaluated with three kinds of corpus,evaluation shows that it can separate all the corpus with excellent performance.