From: A recursive expectation-maximization algorithm for speaker tracking and separation
Overlapping | Speaker | Input | Proposed | ODAS |
---|---|---|---|---|
20% | 1 | 3.44 (0.94) | 3.90 (0.97) | 3.45 (0.83) |
20% | 2 | 2.90 (0.94) | 3.73 (0.98) | 2.89 (0.83) |
40% | 1 | 2.77 (0.88) | 3.63 (0.96) | 2.70 (0.79) |
40% | 2 | 2.38 (0.89) | 2.52 (0.83) | 2.41 (0.80) |
60% | 1 | 2.33 (0.83) | 3.37 (0.95) | 2.42 (0.76) |
60% | 2 | 2.01 (0.84) | 3.20 (0.95) | 2.05 (0.75) |
80% | 1 | 1.87 (0.79) | 3.10 (0.94) | 1.97 (0.74) |
80% | 2 | 1.84 (0.79) | 3.02 (0.94) | 1.90 (0.73) |