基于多基音跟踪的单声道混合语音分离-计算机应用研究.PDF

第 25 卷第 6 期 计 算 机 应 用 研 究 Vol. 25 No. 6 2008 年 6 月 Application Research of Computers Jun. 2008 基于多基音跟踪的单声道混合语音分离 * a b b a, b 李 鹏 , 关 勇 , 刘文举 , 徐 波 ( 中国科学院 自动化研究所 a. 数字媒体内容技术研究中心; b. 模式识别国家重点实验室 , 北京 100080) 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。 关键词: 计算听觉场景分析; 多基音跟踪; 语音分离 中图分类号: TN912. 3 文献标志码: A 文章编号 : 1001- 3695( 2008) 06- 1660- 03 Monaural speech separation based on multi-pitch tracking a b b a, b LI Peng , GUAN Yong , LIU Wen-ju , XU Bo ( a. Digital Media Content Technology Research Center, b. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing 100080, China) Abstract: This paper proposed a multi-pitch tracking based monaural speech separation systemto 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. Key words: computational auditory scene analysis( CASA