Evaluation of Frequency-Domain Blind Signal Separation Using Directivity Pattern Under Reverberant Conditions

WESTPRAC VII 2000: the 7th West Pacific Regional Acoustics Conference, October 3-5, 2000, Kumamoto, Japan.This paper describes a new blind signal separation method using the directivity patterns of the microphone array and its objective evaluation under reverberant conditions. In this method, to deal with the arriving lags among each microphone, the inverses of the mixing matrices are calculated in the frequency domain so that the separated signals are mutually independent. Since the calculations are carried out in each frequency independently, the following problems arise: (1) permutation of each sound source, (2) arbitrariness of each source gain. To resolve these problems, a priori assumption of similarity among the source signal waveforms must be required in previous works. In this paper, we propose a new solution that directivity patterns are explicitly used to estimate the each sound source direction. Using this method, we can resolve both permutation and arbitrariness problems simultaneously without the assumption for the source signal waveforms. As the results of signal separation experiments under various reverberant conditions, it is shown that the proposed method improves the SNR of degraded speech by 17.5 dB under nonreverberant condition. Also, the proposed method improves the SNR by 8.0 dB when the reverberation time is 0.21 sec, and by 4.8 dB when the reverberation time is 0.36 sec; these performances are superior to those of previous method