HUG-VIS - Emotion Recognition (Audio): leaderboard
HUG-VIS (2026) records 30 professional actors performing the same 280 emotion-conditioned assignments. This board scores zero-shot seven-class emotion recognition from audio input over its 8,400 clips, as accuracy in percent; higher is better.
Metric: Seven-class emotion recognition accuracy (%; 8,400 seated half-body clips of 30 professional actors, each performing the same 280 emotion-action-prompt assignments in a controlled Mandarin studio, 1,200 clips per emotion; zero-shot; input: audio). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Audio-Reasoner-7B | 74.38 | |
| 2 | EmotionThinker | 72.98 | |
| 3 | Qwen2-Audio-7B-Instruct | 69.85 | |
| 4 | Kimi-Audio-7B-Instruct | 64.64 | #1433 |
Interactive version: theaggregate.ai/benchmark?slug=hug-vis-emotion-recognition-audio · How It Works · Data refreshed daily, snapshot 2026-11-13.