Speaker Diarization on CallHome 2-spks with 0.25s collar tolerance (test)
9.01DER (%)EEND-GLA-Small+BW-STB
Evaluation Results
| Method | Links | |
|---|---|---|
| EEND-GLA-Small+BW-STBBuffer (s)=100, Latency (s)=12025.12 | 9.01 | |
| EEND-EDA+FW-STB (Improved Training)Buffer (s)=101, Latency (s)=1, Note=Reproduced results with improved training methodology2025.12 | 9.08 | |
| EEND-GLA-Large+BW-STBBuffer (s)=100, Latency (s)=12025.12 | 9.2 | |
| O-EENC-SDBuffer (s)=∞, Latency (s)=52025.12 | 9.33 | |
| O-EENC-SDBuffer (s)=∞, Latency (s)=102025.12 | 9.5 | |
| O-EENC-SDBuffer (s)=100, Latency (s)=102025.12 | 9.5 | |
| O-EENC-SDBuffer (s)=100, Latency (s)=52025.12 | 9.53 | |
| O-EENC-SDBuffer (s)=10, Latency (s)=102025.12 | 10.75 | |
| O-EENC-SD (5s/50s training)Buffer (s)=∞, Latency (s)=1, Training latency=5-s, Training buffer=50-s2025.12 | 11.45 | |
| BW-EDA-EENDBuffer (s)=10, Latency (s)=102025.12 | 11.82 | |
| O-EENC-SD (5s/50s training)Buffer (s)=100, Latency (s)=1, Training latency=5-s, Training buffer=50-s2025.12 | 11.96 | |
| O-EENC-SD (5s/50s training)Buffer (s)=25, Latency (s)=1, Training latency=5-s, Training buffer=50-s2025.12 | 12.14 | |
| O-EENC-SD (Full)Buffer (s)=100, Latency (s)=12025.12 | 12.47 | |
| O-EENC-SD (5s/50s training)Buffer (s)=50, Latency (s)=1, Training latency=5-s, Training buffer=50-s2025.12 | 12.54 | |
| O-EENC-SD Base + Centroid decoderBuffer (s)=100, Latency (s)=12025.12 | 12.69 | |
| EEND-EDA+FW-STBBuffer (s)=101, Latency (s)=12025.12 | 12.7 | |
| O-EENC-SDBuffer (s)=5, Latency (s)=52025.12 | 13.2 | |
| O-EENC-SD (5s/50s training)Buffer (s)=10, Latency (s)=1, Training latency=5-s, Training buffer=50-s2025.12 | 14.5 | |
| O-EENC-SD Base + Attractors decoderBuffer (s)=100, Latency (s)=12025.12 | 15.07 | |
| O-EENC-SD BaseBuffer (s)=100, Latency (s)=12025.12 | 15.38 | |
| O-EENC-SD (5s/50s training)Buffer (s)=5, Latency (s)=1, Training latency=5-s, Training buffer=50-s2025.12 | 19.99 |