DEAF
product→ stable
Diagnostic Evaluation of Acoustic Faithfulness
DEAF is a 2,700-stimulus benchmark created by researchers to test the acoustic processing capabilities of audio-based multimodal large language models (MLLMs).
1Total Mentions
+0.10Sentiment (Neutral)
0.0%Velocity (7d)
First seen: Mar 20, 2026Last active: Mar 20, 2026
Signal Radar
Five-axis snapshot of this entity's footprint
Loading radar…
Mentions × Lab Attention
Weekly mentions (solid) and average article relevance (dotted)
mentionsrelevance
Loading timeline…
Timeline
1- Research MilestoneMar 17, 2026
DEAF benchmark paper published on arXiv revealing Audio MLLMs' reliance on text over acoustic processing
View source- benchmark size:
- 2,700-stimulus
- key finding:
- models scoring below 50% on acoustic faithfulness
Relationships
No relationships mapped yet.
Recent Articles
No articles found for this entity.
Predictions
No predictions linked to this entity.
AI Discoveries
No AI agent discoveries for this entity.
Sentiment History
Positive sentiment
Negative sentiment
Range: -1 to +1
| Week | Avg Sentiment | Mentions |
|---|---|---|
| 2026-W12 | 0.10 | 1 |