Speaker Diarization on CALLHOME
7.2DER (All Speakers)ECAPA-TDNN with EEND-EDA
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ECAPA-TDNN with EEND-EDANumber of speakers=2-4 Speakers, VAD config=Oracle VAD, Complexity=higher modelling and deployment complexity, Input features=Mel-filterbank features2026.06 | 7.2 | — | — | — | — | — | 68.8 | |
| WavLM Large & EEND-vector clusteringEstimated number of speakers=true2021.10 | 10.35 | 6.46 | 10.69 | 11.84 | 12.89 | 20.7 | — | |
| WavLM Base+ & EEND-vector clusteringEstimated number of speakers=true2021.10 | 11.78 | 6.99 | 11.12 | 15.2 | 21.61 | 21.7 | — | |
| EEND-EDA clusteringEstimated number of speakers=true2021.10 | 11.84 | 7.11 | 11.88 | 14.37 | 25.95 | 21.95 | — | |
| HuBERT Large & EEND-vector clusteringEstimated number of speakers=true2021.10 | 12.4 | 7.39 | 11.97 | 15.76 | 19.82 | 22.1 | — | |
| EEND-vector clusteringEstimated number of speakers=true2021.10 | 12.49 | 7.96 | 11.93 | 16.38 | 21.21 | 23.1 | — | |
| HUBERT Base & EEND-vector clusteringEstimated number of speakers=true2021.10 | 12.63 | 7.93 | 12.07 | 15.21 | 19.59 | 23.32 | — | |
| VBxOracle speech segments=true, Source of results=[65]2021.10 | 13.28 | 9.44 | 13.89 | 16.05 | 13.87 | 24.73 | — | |
| EEND-vector clusteringEstimated number of speakers=true, implementation=Ours2021.10 | 13.31 | 7.54 | 12.42 | 18.41 | 26.79 | 27.4 | — | |
| EEND-EDAEstimated number of speakers=true2021.10 | 15.29 | 8.5 | 13.24 | 21.46 | 33.16 | 40.29 | — | |
| SC-EENDEstimated number of speakers=true, Source of results=[65]2021.10 | 15.75 | 9.57 | 14 | 21.14 | 31.07 | 37.06 | — | |
| x-vector clusteringEstimated number of speakers=true2021.10 | 19.43 | 15.45 | 18.01 | 22.68 | 31.4 | 34.27 | — |