Improved Method of Blind Speech Separation with Low Computational Complexity

Abstract EN

A blind speech separation method with low computational complexity is proposed.

This method consists of a combination of independent component analysis with frequency band selection, and a frame-wise spectral soft mask method based on an interchannel power ratio of tentative separated signals in the frequency domain.

The soft mask cancels the transfer function between sources and separated signals.

A theoretical analysis of selection criteria and the soft mask is given.

Performance and effectiveness are evaluated via source separation simulations and a computational estimate, and experimental results show the significantly improved performance of the proposed method.

The segmental signal-to-noise ratio achieves 7 [dB] and 3 [dB], and the cepstral distortion achieves 1 [dB] and 2.5 [dB], in anechoic and reverberant conditions, respectively.

Moreover, computational complexity is reduced by more than 80% compared with unmodified FDICA.