Comparison of the standard three-factor disentanglement methods with re-entry training framework on singing voices.
| Methods | MSE ↓ | Phoneme estimation [%] | Pitch estimation [%] | ||||
|---|---|---|---|---|---|---|---|
| (Timbre ↓) | Variation ↑ | Pitch ↓ | (Timbre ↓) | Variation ↓ | Pitch ↑ | ||
| Standard | 0.613 | (N/A) | 61.4 | 48.1 | (N/A) | 12.7 | 45.5 |
| Standard + RN | 1.130 | (N/A) | 60.8 | 58.5 | (N/A) | 11.9 | 53.2 |
| Standard + PD | 0.594 | (N/A) | 60.9 | 44.8 | (N/A) | 12.5 | 44.0 |
| Standard + RN + PD | 1.110 | (N/A) | 60.7 | 58.2 | (N/A) | 11.7 | 51.1 |
| Re-entry | 0.590 | (N/A) | 60.7 | 33.9 | (N/A) | 10.9 | 33.6 |
| Methods | MSE ↓ | Phoneme estimation [%] | Pitch estimation [%] | ||||
|---|---|---|---|---|---|---|---|
| (Timbre ↓) | Variation ↑ | Pitch ↓ | (Timbre ↓) | Variation ↓ | Pitch ↑ | ||
| Standard | 0.613 | (N/A) | 61.4 | 48.1 | (N/A) | 12.7 | 45.5 |
| Standard + RN | 1.130 | (N/A) | 60.8 | 58.5 | (N/A) | 11.9 | 53.2 |
| Standard + PD | 0.594 | (N/A) | 60.9 | 44.8 | (N/A) | 12.5 | 44.0 |
| Standard + RN + PD | 1.110 | (N/A) | 60.7 | 58.2 | (N/A) | 11.7 | 51.1 |
| Re-entry | 0.590 | (N/A) | 60.7 | 33.9 | (N/A) | 10.9 | 33.6 |