AudioCC Lab · Text-to-Audio Evaluation
PhyAudioBench Leaderboard
Evaluating whether generated sounds are physically consistent, not just semantically plausible. Explore model performance across acoustic and prompt fidelity (L1), event structure (L2), and physical consistency (L3).
Evaluate your model by following the instructions in our
GitHub repository,
and submit your evaluation results to
yuchaojia82@gmail.com
for inclusion in the leaderboard.
Model Rankings
Higher is better · Click a column heading to change the ranking metric
| Rank | Model | L1 ↕ | L2 ↕ | L3 ↕ | Overall ↕ |
|---|
Example results from the existing PhyAudioBench evaluation. Scores are shown as proportions (0–1). Leaderboard submissions are reviewed before publication to ensure consistent evaluation settings and reproducible results.
PhyAudioBench · AudioCC-Lab. To publish updated scores, maintain
leaderboard.json in the public results Dataset.