Table 3 Performance improvement of the proposed network evaluated with the six-channel UCA for different subtending angles
From: A multichannel learning-based approach for sound source separation in reverberant environments
Model | ΔSI-SNR (dB) | ΔPESQ | ΔSTOI | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
0°–15° | 15°–45° | 45°–90° | 90°–180° | Avg. | 0°–15° | 15°–45° | 45°–90° | 90°–180° | Avg. | Avg. | |
BF-net (the first stage) | 7.25 | 7.62 | 7.92 | 8.16 | 7.78 | 0.22 | 0.37 | 0.52 | 0.52 | 0.43 | 0.14 |
BF-net + LSTM | 7.62 | 8.73 | 9.30 | 9.04 | 8.86 | 0.17 | 0.33 | 0.52 | 0.49 | 0.41 | 0.15 |
BF-net + U-net | 12.29 | 12.87 | 12.51 | 12.48 | 12.62 | 0.32 | 0.63 | 0.84 | 0.79 | 0.70 | 0.20 |
BF-net + U-net + DM | 14.32 | 14.27 | 13.69 | 13.81 | 13.99 | 0.51 | 0.78 | 0.95 | 0.89 | 0.83 | 0.22 |
Score of unprocessed signal | − 16.72 | − 13.15 | − 10.93 | − 12.95 | − 12.73 | 1.46 | 1.56 | 1.60 | 1.51 | 1.55 | 0.59 |