基于听觉场景分析的单声道双人混合语音浊音分离  

Voiced Separation of Monophonic Two-person Mixed Speech Based on Auditory Scene Analysis

作　　者：张晗 张二华[1] 姜珊 ZHANG Han;ZHANG Erhua;JIANG Shan(School of Computer Science and Engineering,Nanjing University of Science&Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机与数字工程》2022年第11期2461-2466,2496,共7页Computer & Digital Engineering

基　　金：军委装备部十三五装备预研领域基金项目(编号:61403120102)资助。

摘　　要：语言是人类最重要的信息交流方式,在实际环境中,各种噪声严重影响语音的音质及清晰度,使语音识别的正确率明显下降。若应用语音分离技术将目标语音从混合语音中分离出来,则能有效提高语音识别的性能。语音分离可分为单声道语音分离和多声道语音分离,其中单声道语音分离最为困难。论文研究了基于听觉场景分析(CASA)的单声道语音分离方法,首先以基音周期轨迹为线索,根据浊音的谐波结构,分离出目标语音的频谱,再由傅里叶反变换(IFFT)重构分离语音,针对分离语音中存在的窜音现象,采用振幅平滑和相位调整法消除窜音。实验结果表明,该方法能有效消除窜音现象,显著提高分离语音的清晰度。Language is the most important way of communicating information for human beings.In the actual environment,various noises seriously affect the sound quality and clarity of speech,which significantly reduces the accuracy of speech recognition.If the speech separation technology is used to separate the target speech from the mixed speech,the performance of speech recognition can be effectively improved.Speech separation can be divided into monophonic speech separation and multi-channel speech separation,among which monophonic speech separation is the most difficult.This paper studies the monophonic speech separation method based on auditory scene analysis(CASA).First,the pitch cycle track is used as a clue to separate the frequency spectrum of the target speech according to the harmonic structure of the voiced sound,and then the inverse Fourier transform(IFFT)Reconstruct the separated speech,aiming at the crosstalk phenomenon in the separated speech,adopt amplitude smoothing and phase adjustment to eliminate the crosstalk.Experimental results show that this method can effectively eliminate crossover phenomenon and significantly improve the clarity of separated speech.

关 键 词：语音分离 基音周期 窜音 振幅平滑 相位调整法 

分 类 号：TM912[电气工程—电力电子与电力传动]