Active Speaker Detection Inference Efficiency Profiling
1.25VRAM (GB)LR-ASD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LR-ASDVideo frames=10002025.12 | 1.25 | 38.63 | 25,890 | |
| TalkNetVideo frames=10002025.12 | 1.89 | 48.65 | 20,556 | |
| LR-ASDVideo frames=20002025.12 | 2.5 | 78.64 | 25,431 | |
| GateFusionVideo frames=10002025.12 | 3.35 | 77.2 | 12,953 | |
| TalkNetVideo frames=20002025.12 | 3.52 | 102.52 | 19,508 | |
| GateFusionVideo frames=20002025.12 | 4.96 | 174.37 | 11,470 | |
| LR-ASDVideo frames=40002025.12 | 4.99 | 156.09 | 25,627 | |
| LoCoNetVideo frames=10002025.12 | 5.02 | 135.45 | 7,383 | |
| TalkNetVideo frames=40002025.12 | 6.96 | 213.88 | 18,702 | |
| LR-ASDVideo frames=60002025.12 | 7.48 | 236.32 | 25,390 | |
| GateFusionVideo frames=40002025.12 | 8.17 | 423.37 | 9,448 | |
| LoCoNetVideo frames=20002025.12 | 10.19 | 274.97 | 7,273 | |
| TalkNetVideo frames=60002025.12 | 10.32 | 328.01 | 18,292 | |
| GateFusionVideo frames=60002025.12 | 11.37 | 749.56 | 8,005 |