Speaker Diarization on AISHELL-4
1.6DER (%)DiariZen + Oracle #Spk
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DiariZen + Oracle #SpkClustering=Oracle2026.01 | 1.6 | 4.8 | 1.2 | |
| Hyper-SDCollar (ζ)=0.25s, Space=Hyperbolic2026.01 | 4.44 | — | — | |
| DiarizenCollar (ζ)=0.25s, Space=Euclidean2026.01 | 4.78 | — | — | |
| Pyannote3Collar (ζ)=0.25s, Space=Euclidean2026.01 | 6.27 | — | — | |
| Multi-channel DiariZenClustering=Oracle2026.01 | 8.9 | — | — | |
| DiariZen + Spatial DiarizationClustering=Oracle2026.01 | 8.9 | 18.5 | 7.8 | |
| DiariZen + Spatial Diarization + Joint FinetuningClustering=Oracle2026.01 | 8.9 | 17.4 | 8 | |
| DiariZenClustering=Oracle2026.01 | 9.1 | 16.4 | 8.3 | |
| DiariZen + Spatial ConformerClustering=Oracle2026.01 | 9.4 | 18.4 | 8.5 | |
| DiariZen + Spatial EncoderClustering=Oracle2026.01 | 9.5 | 17.9 | 8.6 | |
| Hyper-SDCollar (ζ)=0s, Space=Hyperbolic2026.01 | 9.52 | — | — | |
| DiariZen-Large ConformerClustering=Oracle2026.01 | 9.6 | 16 | 8.9 | |
| Dual-Strategy-Enhanced ConBiMambacollar=02026.01 | 9.8 | — | — | |
| DiarizenCollar (ζ)=0s, Space=Euclidean2026.01 | 9.94 | — | — | |
| Spatial DiarizationClustering=Oracle2026.01 | 10 | 22.5 | 8.6 | |
| Mamba-diarizationcollar=02026.01 | 10.5 | — | — | |
| Diarizen(WavLM-frozen)collar=02026.01 | 11.7 | — | — | |
| Diarizen(WavLM-updated)collar=02026.01 | 11.7 | — | — | |
| PyannoteAIcollar=02026.01 | 11.9 | — | — | |
| Pyannote3Collar (ζ)=0s, Space=Euclidean2026.01 | 11.96 | — | — | |
| VBxClustering=VBx [24]2026.06 | 15.8 | — | — | |
| SphereVBx-PFClustering=SphereVBx-PF2026.06 | 16 | — | — | |
| SphereVBxClustering=SphereVBx2026.06 | 16.1 | — | — | |
| VBxClustering=VBx (ours)2026.06 | 16.7 | — | — |