2010 | OriginalPaper | Chapter
Frequency-Domain Blind Separation of Convolutive Speech Mixtures with Energy Correlation-Based Permutation Correction
Authors : Li-Dan Wang, Qiu-Hua Lin
Published in: Advances in Neural Network Research and Applications
Publisher: Springer Berlin Heidelberg
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
Blind separation of convolutive speech mixtures in frequency domain has obvious advantages in term of convergence and computation, but suffers from permutation ambiguity. Motivated by the fact that speech signals have strong correlations across frequency, the paper presents an energy correlation method for solving permutation ambiguity after separation of instantaneous speech mixtures at each frequency bin. Extensive experiments with synthetic and recorded speech signals are carried out to compare the energy correlation method to amplitude correlation method, three different complex-valued independent component analysis (ICA) algorithms are compared as well. The results show that the proposed method achieves better performance than the amplitude correlation method, and the complex ICA algorithm based on negentropy maximization yields the best separation.