《基于人耳听觉系统模型的多声源定位与语音分离研究》由会员分享,可在线阅读,更多相关《基于人耳听觉系统模型的多声源定位与语音分离研究(70页珍藏版)》请在金锄头文库上搜索。
1、河北工业大学硕士学位论文基于人耳听觉系统模型的多声源定位与语音分离研究姓名:武方申请学位级别:硕士专业:机械工程指导教师:戴士杰20091201? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?i?THE RESEARCH OF MULTI SOUND SOURCES LOCALIZATION AND SPEECH SEPARATION BASED ON THE MODEL OF HUMAN AUDITORY SYSTEM ABSTRACTThe technology of sound source localization and separation is
2、part of signal processing, and it has a wide range of application areas and application requirements, such as the study of humanoid robot, the military field, voice communication, target tracking, speaker recognition software front-end pre-processing, strong voice for noisy environment, cahier in la
3、rge areas, hearing aids, and so on. For the status of current technology of sound source localization and separation, studied the physiology and psychological characteristics of human auditory system, and then established a sound source localization model and a speech separation model. In the sound
4、source localization studies, studied the statistical characteristic of time interval between the zero-crossing points, and concluded that the complex signals has the same characteristic as its frequency components part. Base on this conclusion, proposed a hypothesis that using upward-going zero-cros
5、sing points to simulate fire of nerves, then, introduced a time delay algorithm based on time interval of zero-crossing points, and then, resolved the problem of time confusion in high frequency band. Compared with cross-correlation algorithm, this algorithm has fewer calculation amounts, stronger a
6、bility of anti-noise and more accurate multi-source localization. Speech separation is based on the localization. In the algorithm, used the signal amplitude at zero-crossing points as perception element, and allocated these amplitudes to corresponding sound sources according to the time delay, and
7、then reconstructed the sound after time scale in every frequency band. This algorithm simulates the feature extraction, classification and reconstruction process of the human ear. The simulation experiment of speech separation proved that the algorithm can extracting speech signal from noise environ
8、ment and separating the mixed speech signals. Analyzed the real sound environment, and for the problems such as noise, reverberation and continuity, introduced cochlear filter and signal endpoint detection. Built a real-time voice capture and sound source localization system. Experimental results sh
9、owed that the algorithm achieved good results at low SNR, multi-source cases. KEY WORDS: sound source localization, speech separation, human auditory, upward-going zero-crossings points, ii? ? ? ? ? ? ? ? ? ? ? ? ? ?1-1? ? 1876 ? ?1?1-2 ?2, 3? 2006 ?“RI-MAN”? 1.1 ?TIME?“RI-MAN”?“?”?1 ? 1.1 ?RI-MAN?Fig. 1.1 The “RI-MAN” partner robot designed by the RIKEN Bio-Mimetic Control Research Center. ?