Table 1 Cross-domain dataset information
From: Heterogeneous separation consistency training for adaptation of unsupervised speech separation
Dataset | Acoustics | Type | #Spks | #Utts | Hours | Oracle |
|---|---|---|---|---|---|---|
Libri2Mix (Source) | English/A | Train | 251 | 13,900 | 58 | \(\checkmark\) |
Dev | 40 | 3000 | 11 | \(\checkmark\) | ||
Test | 40 | 3000 | 11 | \(\checkmark\) | ||
WHAMR! (Target) | English/R | Train | \(101^*\) | 20,000 | 30 | - |
Dev | \(101^*\) | 5000 | 10 | - | ||
Test | 18 | 3000 | 5 | \(\checkmark\) | ||
Aishell2Mix (Target) | Mandarin/A | Train | 340 | 10,000 | 11 | - |
Dev | 40 | 3000 | 3 | - | ||
Test | 20 | 3000 | 3 | \(\checkmark\) |