Table 5 Performance of two-factor and...

Table 5

Performance of two-factor and three-factor disentanglement on singing voices across different split sets.

Data split	Factors			MSE ↓	Phoneme estimation [%]		Pitch estimation [%]
Data split	Timbre	Variation	Pitch	MSE ↓	Variation ↑	Pitch ↓	Variation ↓	Pitch ↑
Test	✓		✓	0.873	—	64.7	—	16.6
Test		✓	✓	0.653	62.1	46.3	13.3	44.2
Test	✓	✓	✓	0.613	61.4	48.1	12.7	45.5
Training	✓		✓	0.897	—	63.8	—	14.9
Training		✓	✓	0.650	62.4	44.4	11.5	43.0
Training	✓	✓	✓	0.605	61.5	44.2	10.7	45.2
Validation	✓		✓	0.888	—	63.2	—	17.8
Validation		✓	✓	0.657	60.9	44.0	12.8	45.0
Validation	✓	✓	✓	0.614	60.3	46.2	11.1	47.3

Data split	Factors			MSE ↓	Phoneme estimation [%]		Pitch estimation [%]
Data split	Timbre	Variation	Pitch	MSE ↓	Variation ↑	Pitch ↓	Variation ↓	Pitch ↑
Test	✓		✓	0.873	—	64.7	—	16.6
Test		✓	✓	0.653	62.1	46.3	13.3	44.2
Test	✓	✓	✓	0.613	61.4	48.1	12.7	45.5
Training	✓		✓	0.897	—	63.8	—	14.9
Training		✓	✓	0.650	62.4	44.4	11.5	43.0
Training	✓	✓	✓	0.605	61.5	44.2	10.7	45.2
Validation	✓		✓	0.888	—	63.2	—	17.8
Validation		✓	✓	0.657	60.9	44.0	12.8	45.0
Validation	✓	✓	✓	0.614	60.3	46.2	11.1	47.3