Table 10 Results using a single mfcc stream and all the proposed methods
From: Analysis of transition cost and model parameters in speaker diarization for meetings
System | SER DEVELSET (± 0.04) | Relative improvement over baseline | SER RT09 (± 0.09) | Relative improvement over baseline |
|---|---|---|---|---|
Baseline | 8.98% | 21.21% | ||
Transition weight (transition weight, minimum duration) (0.001-350) | 8.29 | 8.3% | 14.87% | 42% |
Model selection (A, B, minimum duration) (12-30-350) | 8.14% | 10.31% | 18.34% | 15.64% |
Model selection (A, B, minimum duration) (11-40-350) | 8.3% | 8.19% | 15.51% | 36.75% |
Fusion (transition weight, A,B, minimum duration) (0.001-11-30-350) | 8.22% | 9.2% | 20.62% | 2.8% |
Fusion (transition weight, A,B, minimum duration) (0.01-11-30-350) | 8.4% | 6.9% | 18.70% | 13.4% |
Fusion (transition weight, A, B, minimum duration) (0.001-11-40-350) | 8.44% | 6.4% | 16.64% | 27.46% |