基于多基音跟踪的单声道混合语音分离  

Monaural speech separation based on multi-pitch tracking

作　　者：李鹏[1] 关勇[2] 刘文举[2] 徐波[1] 

机构地区：[1]中国科学院自动化研究所数字媒体内容技术研究中心,北京100080 [2]中国科学院自动化研究所模式识别国家重点实验室,北京100080

出　　处：《计算机应用研究》2008年第6期1660-1662,共3页Application Research of Computers

基　　金：国家“973”计划资助项目(2004CB318105)

摘　　要：针对许多计算听觉场景分析系统无法很好地解决多说话人混合语音信号分离的问题,提出了一种基于多基音跟踪的单声道混合语音分离系统。该系统充分利用了多基音跟踪研究的最新成果,通过将多基音跟踪得到的目标语音和干扰语音的基音轨迹信息结合到分离系统中,有效地改善了分离系统在包括多说话人混合在内的多种干扰情况下的分离效果,为多说话人语音分离问题的解决提供了新的思路。This paper proposed a multi-pitch tracking based monaural speech separation system to solve the problem that many computational auditory scene analysis（CASA） systems could not process the mixture signal of multi-speaker.The proposed system made use of the latest achievement in the field of multi-pitch tracking.Through combing the pitch contours of target speech and interference acquired by multi-pitch tracking algorithm into the CASA system,improved obviously the performance of separation system for various mixture conditions,including two-speaker mixture condition,which offered a good solution to separate the mixture speech of multi-speaker.

关 键 词：计算听觉场景分析 多基音跟踪 语音分离 

分 类 号：TN912.3[电子电信—通信与信息系统]