基于计算听觉场景分析的单声道浊音分离  

Monaural voiced speech separation based on computational auditory scene analysis

作　　者：张丽娜 张二华 江军亮 ZHANG Li-na;ZHANG Er-hua;JIANG Jun-liang(School of Computer Science and Engineering,Nanjing University of Science & Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学计算机科学与工程学院

出　　处：《计算机工程与科学》2019年第7期1266-1272,共7页Computer Engineering & Science

基　　金：军委装备发展部“十三五”装备预研领域基金(61403120102)

摘　　要：针对单声道语音分离中浊音分离的问题,提出了一种准确估计基音周期的方法。首先,以语音的短时平稳性和基音周期的连续性等为线索,利用语音信号的倒谱峰值构成基音周期谱图,并自动提取基音周期轨迹。然后,利用谐波频率为基音频率整数倍的性质来拾取各次谐波的频谱。最后,通过傅里叶逆变换对浊音进行重构。实验结果表明,该方法能准确提取基音周期轨迹,有效分离浊音信号。Aiming at the problem of voiced speech separation in monaural speech separation,we propose an accurate pitch period estimation method.Firstly,using the short-term stability of speech and the continuity of the pitch period as clues,we use the cepstrum peak of speech signals to form the pitch spectrum,and the pitch period track is automatically extracted.Then,the spectrum of each harmonic is picked up by using the property that the harmonic frequency is an integer multiple of the fundamental frequency.Finally,the voiced speech is reconstructed by the inverse Fourier transform.Experimental results show that this method can accurately extract the pitch period track and effectively separate voiced signals.

关 键 词：计算听觉场景分析 语音分离 基音周期轨迹 浊音 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]