TABLE VI.
SDR Comparison Between Speech Separation Methods With Target Dependent Training on IEEE-TIMIT Corpus
| −12 dB | −9 dB | −6 dB | −3 dB | O dB | 3 dB | 6 dB | |
|---|---|---|---|---|---|---|---|
| Noisy | −10.86 | −8.32 | −5.59 | −2.74 | 0.18 | 3.14 | 6.12 |
| DNN+Map | 2.61 | 4.09 | 5.49 | 6.97 | 8.61 | 10.32 | 12.05 |
| DNN+IRM | 2.48 | 4.21 | 5.89 | 7.61 | 9.45 | 11.32 | 13.16 |
| MCA+IRM | 2.92 | 4.61 | 6.27 | 7.97 | 9.77 | 11.61 | 13.41 |
| MCS+IRM | 3.71 | 5.53 | 7.24 | 8.92 | 10.67 | 12.44 | 14.12 |
| DNN+SA | 4.20 | 5.47 | 6.98 | 8.47 | 10.14 | 11.92 | 13.71 |
| MCA+SA | 4.54 | 5.80 | 7.33 | 8.84 | 10.49 | 12.24 | 14.01 |
| MCS+SA | 4.75 | 6.09 | 7.49 | 9.05 | 10.74 | 12.53 | 14.32 |
| MCA+IRM+SA | 4.31 | 5.71 | 7.16 | 8.73 | 10.43 | 12.21 | 13.98 |
| MCS+IRM+SA | 4.79 | 6.13 | 7.54 | 9.10 | 10.80 | 12.61 | 14.43 |