MuVAP Fisher speaker-mono - Zero Shot Balanced Accuracy: leaderboard
MuVAP Fisher speaker-mono - Zero Shot Balanced Accuracy: Balanced accuracy (0-1). MuVAP official released modules: eleven named VAP checkpoints, epoch4/4, source-native speaker/role/mono codebooks, CPC/Mimi frontends and frame rates. Fisher turn events zero-shot macro F1 and balanced accuracy read from p_future trailing1.4s, no probe/tuning. Codebook-specific shipped shift scale speaker1/role2 (threshold0.5/0.333) retained; test-swept scales withheld. Logistic last-frame probe F1 is separate. Two released ASD frontend variants report AVA-ActiveSpeaker mAP. Bare AVCC MuVAP/RoleVAP scores remain held for exact artifact binding, not because fitted readouts are ineligible.
Metric: Balanced accuracy (0-1). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MuVAP vap-speaker-mono-mimi | 0.72 | |
| 2 | MuVAP vap-speaker-mono-mimi-12hz | 0.72 | |
| 3 | MuVAP vap-speaker-mono-cpc-50hz | 0.68 |
Interactive version: theaggregate.ai/benchmark?slug=muvap-fisher-speaker-mono-zero-shot-balanced-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-12.