摘要
在语音信号频域盲分离算法中,利用短时傅里叶变换将时域卷积变换到频域乘积,从而在每个子频段上进行复数瞬时分离,降低运算量,但这会导致分离结果出现次序和幅度上的不确定性。文章提出了一种改进的基于三阶累积量的频域排序算法,在每个子频段上计算信号的三阶累积量,比较其相关性来解决排序问题。仿真实验证明了该算法的可行性。
For blind source separation of speech signal,the convolution mixture in time-domain is transformed into instantaneous mixture in frequency-domain by using short-time Fourier transform.So complex independent component analysis is performed in each frequency band,but this will bring about the problem of permutation and scaling arbitrary.A frequency domain permutation algorithm based on three-order cumulant is proposed in this paper,in which the cumulant is computed in each frequency band firstly,and then the relevance ratio is compared to solve the permutation problem.The simulation results verify the validity of the proposed algorithm.
出处
《空间电子技术》
2012年第1期11-16,共6页
Space Electronic Technology
基金
部级预研基金资助项目(9140A220309090C0201)
关键词
卷积盲分离
复数JADE
三阶累积量
排序
Convolution blind source separation
Comptex JADE
Three-order cumulant
Permutation
