语音和背景音乐分离算法及其系统设计  

Separation Algorithm and System Design of Speech and Background Music

作　　者：王振宇[1] WANG Zhen-yu(Baoji University Of Artsand Sciences,Baoji 721013 China)

机构地区：[1]宝鸡文理学院

出　　处：《自动化技术与应用》2019年第8期77-79,共3页Techniques of Automation and Applications

摘　　要：语音信号得分离对于解决某些语音识别问题至关重要,该问题的本质是对语音信号进行分离,可通过盲源分离技术进行不同信号的分离。在实际的语音环境中,通常背景音乐与语音混为一体,因此从背景嘈杂的语音中较难分辨出清晰的原声信息。本文提出从混合语音中提取一个或多个语音信号,并利用小波对混合语音信号进行降噪处理,采用ICA和二进制T-F掩蔽法等技术对小波输出混合语音进行源分离,结果证明该方法对于混响条件下的语音分离有较好的效果。The separation of speech signal is very important to solve some problems of speech recognition.The essence of this problem is to separate speech signal,which can be separated by blind source separation technology.In the actual speech environment,background music is usually mixed with voice,so it is difficult to distinguish clear acoustic information from background noisy voice.In this paper,one or more speech signals are extracted from mixed speech,and the mixed speech signals are denoised by wavelet.ICA and binary t-f masking method are adopted to separate the source of the mixed speech output by wavelet.

关 键 词：语音信号分离 盲源分离技术 离散时间小波变换 独量分类分析 

分 类 号：TN911.7[电子电信—通信与信息系统]