Upload folder using huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,17 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# JAME Dataset
|
| 2 |
|
| 3 |
**JAME** (JAM Evaluation) is a comprehensive music dataset containing 250 high-quality music tracks designed for **standardized evaluation of song generation models**.
|
| 4 |
|
| 5 |
This dataset is part of **Project Jamify** developed by [DeCLaRe Lab](https://github.com/declare-lab/jamify/tree/main) and supports research in controllable music generation.
|
| 6 |
|
| 7 |
-
This dataset supports the research described in:
|
| 8 |
-
|
| 9 |
-
**JAM: A Tiny Flow-based Song Generator with Fine-grained Controllability and Aesthetic Alignment**
|
| 10 |
-
*Renhang Liu, Chia-Yu Hung, Navonil Majumder, Taylor Gautreaux, Amir Ali Bagherzadeh, Chuan Li, Dorien Herremans, Soujanya Poria*
|
| 11 |
-
arXiv:2507.20880 (2025)
|
| 12 |
-
🔗 [Paper](https://arxiv.org/abs/2507.20880) | 🔗 [Code](https://github.com/declare-lab/jamify) | 🔗 [Project](https://declare-lab.github.io/jamify)
|
| 13 |
-
|
| 14 |
-
|
| 15 |
## Dataset Overview
|
| 16 |
|
| 17 |
- **Total Tracks**: 250 carefully curated tracks (50 per genre, 5 genres total)
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
library_name: diffusers
|
| 5 |
+
metrics:
|
| 6 |
+
- PER
|
| 7 |
+
- WER
|
| 8 |
+
- SongEval
|
| 9 |
+
- Audio Aesthetics
|
| 10 |
+
- MuQ
|
| 11 |
+
- FAD
|
| 12 |
+
pipeline_tag: text-to-audio
|
| 13 |
+
tags:
|
| 14 |
+
- music
|
| 15 |
+
- art
|
| 16 |
+
- song-generation
|
| 17 |
+
- lyrics-to-song
|
| 18 |
+
- flow-matching
|
| 19 |
+
- direct-preference-optimization
|
| 20 |
+
license: other
|
| 21 |
+
---
|
| 22 |
+
|
| 23 |
+
<div align="center">
|
| 24 |
+
<img src="https://declare-lab.github.io/jamify-logo-new.png" width="200"/ >
|
| 25 |
+
<br/>
|
| 26 |
+
<h1>JAM: A Tiny Flow-based Song Generator with Fine-grained Controllability and Aesthetic Alignment</h1>
|
| 27 |
+
<br/>
|
| 28 |
+
|
| 29 |
+
[](https://arxiv.org/abs/2507.20880) [](https://github.com/declare-lab/jamify) [](https://huggingface.co/declare-lab/JAM-0.5) [](https://huggingface.co/datasets/declare-lab/JAME) [](https://declare-lab.github.io/jamify)
|
| 30 |
+
|
| 31 |
+
</div>
|
| 32 |
+
|
| 33 |
+
|
| 34 |
# JAME Dataset
|
| 35 |
|
| 36 |
**JAME** (JAM Evaluation) is a comprehensive music dataset containing 250 high-quality music tracks designed for **standardized evaluation of song generation models**.
|
| 37 |
|
| 38 |
This dataset is part of **Project Jamify** developed by [DeCLaRe Lab](https://github.com/declare-lab/jamify/tree/main) and supports research in controllable music generation.
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
## Dataset Overview
|
| 41 |
|
| 42 |
- **Total Tracks**: 250 carefully curated tracks (50 per genre, 5 genres total)
|