Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- audio
|
| 5 |
+
- music
|
| 6 |
+
- reward-model
|
| 7 |
+
- audio-language-model
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# MuseCritic
|
| 11 |
+
|
| 12 |
+
MuseCritic is introduced in **MuseCritic: Learning Multi-Aspect Song Rewards
|
| 13 |
+
through Natural-Language Aesthetic Critiques**.
|
| 14 |
+
|
| 15 |
+
MuseCritic is a semi-scalar reward model for long-form song evaluation. Given
|
| 16 |
+
a song and an aesthetic evaluation rubric, the model first generates a
|
| 17 |
+
natural-language critique covering five dimensions and then predicts five
|
| 18 |
+
continuous scores in the interval `(1, 5)`:
|
| 19 |
+
|
| 20 |
+
- Overall coherence
|
| 21 |
+
- Musicality
|
| 22 |
+
- Memorability
|
| 23 |
+
- Structural clarity
|
| 24 |
+
- Vocal naturalness
|
| 25 |
+
|
| 26 |
+
## Resources
|
| 27 |
+
|
| 28 |
+
- **Code:** [WuqnEl/MuseCritic](https://github.com/WuqnEl/MuseCritic)
|
| 29 |
+
- **Paper:** arXiv link coming soon
|
| 30 |
+
|
| 31 |
+
Please refer to the GitHub repository for installation, inference, training,
|
| 32 |
+
and evaluation instructions.
|
| 33 |
+
|