A multichannel learning-based approach for sound source separation in reverberant environments

EURASIP Journal on Audio, Speech, and Music Processing

Table 4 Performance improvement of the proposed network evaluated with the six-channel UCA for different reverberation time

Model	ΔSI-SNR (dB)				ΔPESQ				ΔSTOI
Model	0.16 s	0.36 s	0.61 s	0.9 s	0.16 s	0.36 s	0.61 s	0.9 s	Avg.
BF-net (the first stage)	7.65	7.69	8.10	7.69	0.62	0.46	0.36	0.27	0.14
BF-net + LSTM	9.33	9.70	8.90	7.41	0.63	0.48	0.31	0.18	0.15
BF-net + U-net	11.27	12.48	13.44	13.38	0.95	0.77	0.58	0.48	0.20
BF-net + U-net + DM	12.33	13.78	15.02	14.98	1.10	0.90	0.70	0.59	0.22
Score of unprocessed signal	− 5.67	− 11.45	− 16.17	− 18.27	1.80	1.59	1.43	1.36	0.59