改进的语音频域盲分离算法  

Improved Algorithm of Speech Blind Source Separation in Frequency Domain

作　　者：王霞[1] 尹晶凡 王光艳[2] 张艳[1] 

机构地区：[1]河北工业大学电子信息工程学院,天津300401 [2]天津商业大学信息工程学院,天津300134

出　　处：《科学技术与工程》2017年第4期87-91,共5页Science Technology and Engineering

基　　金：天津市自然科学基金重点项目(14JCZDJC32600)资助

摘　　要：针对语音信号频域盲分离的排序问题,研究了幅值函数和频点间距对信号相关性的影响,提出了利用短时平均幅度函数和加入频点系数的方法对算法进行改进。改进算法考虑了幅值函数的稳定性和远距离频点的不可靠性,最终实现全部频点的排序。仿真实验对两种不同情况下混合的语音信号进行分离,结果表明改进算法的有效性。For the permutation problem of frequency domain speech blind source separation, the influence of amplitude function and distance between frequency bin on correlation were studied. An improved method based on short-time average amplitude function and coefficient of frequency bin was proposed. It considered the stability of amplitude function and unreliability of frequency bin for long distance, finally realize the process of permuta-tion. Simulation experiment separate speech signal for two different mixed conditions, the results demonstrate the proposed algorithm is good.

关 键 词：频域盲分离 排序 相关函数 频点系数 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]