基于计算听觉场景分析的单通道语音分离方法  

Single-channel Speech Separation Method Based on Computational Auditory Scene Analysis

作　　者：徐庆达 张二华[1] XU Qingda;ZHANG Erhua(School of Computer Science and Engineering,Nanjing University of Science and 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3期597-602,共6页Computer & Digital Engineering

摘　　要：人耳听觉系统能够从嘈杂的环境中筛选出自己感兴趣的语音,基于计算听觉场景分析的方法,论文采用倒谱法提取语音基音周期轨迹,以连续的基音周期轨迹为线索,按基音频率的整数倍提取各次谐波的频谱,再通过傅里叶逆变换重构分离后的语音。实验表明,在几种典型噪音环境下,该方法能有效将目标语音从背景噪声中分离,信噪比(SNR)和评价意见分(MOS)得到一定的提升,平均增益分别为5.67dB和0.36。The human ear hearing system can filter out the speech of interest from the noisy environment.Based on the method of computing auditory scene analysis,this paper uses cepstrum method to extract the pitch trajectory of the speech,and uses the continuous pitch trajectory as a clue.The spectrum of each harmonic is extracted by integer multiples,and the separated speech is reconstructed by inverse Fourier transform.Experiments show that the method can effectively separate the target speech from the background noise in several typical noise environments,and the signal-to-noise ratio(SNR)and mean opinion score(MOS)are improved to a certain extent,with average gains of 5.67dB and 0.36 respectively.

关 键 词：听觉场景分析 语音分离 基音周期 

分 类 号：TN912[电子电信—通信与信息系统]