UDIO-24M

<h1 align='center'>Udio 24M</h1>
<div align="center">
<img src="udio.gif" alt="loading", width="100">
</div>

Excited to introduce, UDIO-24M, the largest known collection of artificial intelligence generated songs alongside SUNO-1M. Both of them are contributions under Sleeping-Imagination initiative of Sleeping AI.

### What's Sleeping-Imagination?
It is an continuous effort where we plan to release high-quality data for artificial intelligence songs and we want to become the largest contributor in this subfield of Generative Music Modelling. Its an ever expanding project, where we continuously update and add more datasets.

### Will there be a paper?
We plan to drop an arXiv paper somewhere in coming weeks and add additional features and data-points for training models on these datasets a breeze under Sleeping-Imagination name. And we also have a planned paper for ICLR 2026.
### Why UDIO?
Udio is one of the Ivy league flavoured AI song generators in the field, which creates sleek and charming songs Vs SUNO's mundane flavour. Quality and features of Udio's songs are quite similar to those of DJ, party and electrified dystopia.

It was only natural that Sleeping AI would create a Udio dataset.

### Mechanism of downloading?
That's something we can't share, even if I want to open-source all my code. We also don't share the usernames to retain competitive advantage.

### Will there be updates and more songs be added?
Currently, we believe to have extracted entire Udio library. Maybe in future, if we have funding and resources we'll update this dataset.
### How many songs and metadata we're talking?
23.4 million songs to be exact and we actually have extracted all possible metadata but due to privacy and competitive reasons we don't share them in public. But, we do provide hyperlinks for both audio and video for individuals to download these songs.
### Can we be taken down in DMCA?
Probably yah! But, Udio also trained on someone else's data. and we don't share any data whatsoever except only hyperlinks.
### What're the data fields?

1. `uuid`: each song has an unique identifier
2. `user_id`: tracing each user
3. `id`: song ids
4. `original_song_url`: downloading the audio
5. `video_path`: LINK to video
### Ethics statement
We responsibly scraped the data according to EU zone and local laws for scientific and research purpose.
### LICENCE
Sleeping AI only permits the data to be used for research and non-commercial purpose under CC-by-nc-nd 4.0. It is to both protect ourselves and be responsible researchers.

<h1 align="center">Udio 24M</h1>
<div align="center">
<img src="udio.gif" alt="加载中", width="100">
</div>

<p>在此隆重推出UDIO-24M——目前已知规模最大的人工智能生成音乐合集，与SUNO-1M一同作为Sleeping AI旗下Sleeping-Imagination项目的贡献成果。</p>

### 何为Sleeping-Imagination项目？
<p>本项目为一项长期推进的计划，旨在发布高质量人工智能生成音乐数据集，并致力于成为生成式音乐建模（Generative Music Modelling）这一子领域内最大的数据集贡献方。本项目处于持续扩张状态，我们将不断更新并新增更多数据集。</p>

### 是否会发表相关论文？
<p>我们计划于未来几周在arXiv平台发布相关研究论文，并以Sleeping-Imagination的名义推出更多可轻松用于模型训练的特征与数据点，同时我们还筹备了一篇将投稿至ICLR 2026的论文。</p>

### 为何选择Udio数据集？
<p>Udio是当前领域内具备常春藤级水准的AI音乐生成工具之一，其生成的音乐风格精致灵动，与SUNO的平实风格形成鲜明对比。Udio生成音乐的音质与特色，与电子舞曲、派对场景音乐以及充满电子感的反乌托邦风格音乐颇为相近。因此，Sleeping AI推出Udio数据集实为顺理成章之举。</p>

### 下载方式如何？
<p>此部分内容暂无法公开，即便我们有意将全部代码开源。为保留竞争优势，我们也不会披露相关用户账号信息。</p>

### 是否会更新数据集并新增更多音乐？
<p>目前我们已完成Udio全曲库的提取工作。未来若能获得足够资金与资源，我们将对本数据集进行更新。</p>

### 本数据集包含多少首音乐与元数据？
<p>确切而言，本数据集包含2340万首AI生成音乐。我们已提取全部可用元数据，但出于隐私保护与竞争考量，暂不公开元数据内容。不过我们为用户提供了音乐音频与视频的超链接，供个人自行下载。</p>

### 是否会因数字千年版权法案（DMCA）要求而被下架？
<p>存在此种可能。不过Udio模型本身亦基于他人数据训练，且我们仅提供超链接，不直接共享任何数据集内容。</p>

### 数据集包含哪些数据字段？
<ol>
<li><code>uuid</code>: 每首音乐均配有唯一标识符</li>
<li><code>user_id</code>: 用于追踪用户身份</li>
<li><code>id</code>: 音乐ID</li>
<li><code>original_song_url</code>: 音频下载链接</li>
<li><code>video_path</code>: 视频资源链接</li>
</ol>

### 伦理声明
<p>本数据集严格遵循欧盟及当地法律法规，以科学研究为目的合规爬取所得。</p>

### 授权协议
<p>Sleeping AI仅允许本数据集以CC-by-nc-nd 4.0协议用于科研与非商业用途，此举既是为了保护我方权益，亦是为了恪守科研人员的责任准则。</p>