Riffusion-X1

Introducing riffusion, it is the last commercial artificial intelligence oriented song generator player in the market and also one of the high-quality early players. when it was unveiled was touted as the stable diffusion for music. And it was also used in one of the earliest known **fake song benchmark** that came out from Mila lab.

We are releasing 99,293 songs and those songs exquisite metadata alongside for fuelling state-of-art research in this field. We also request everyone to obey our dataset terms which are as:

1. In CC-by-nc-nd, we don't allow the commercial use of this dataset
2. We also prohibit any derivative of this dataset from being reuploaded/uploaded anywhere on the internet. Including the sharing of this dataset.

Respect the rules for helping us and everyone to publish high-quality research datasets legally.

> A big shout-out to Azeem for creating this dataset.

## Metadata Field Descriptions

| Field | Description |
|--------------------|-----------------------------------------------------------------------------|
| `id` | Unique identifier for the generated content. |
| `group_id` | ID linking related variations or remixes. |
| `author_id` | User or system ID that created the content. |
| `created_at` | Timestamp of when the content was generated. |
| `audio_variation` | Type or version of audio variation (e.g., original, remix). |
| `lyrics` | Raw lyrics (non-timestamped) associated with the audio. |
| `title` | Title of the generated piece. |
| `image_override` | Flag or ID indicating if custom image was used. |
| `sound` | Possibly a tag/category of the sound style. |
| `topic` | Theme/topic associated with the generation. |
| `image_id` | ID of the generated or attached image. |
| `simple_waveform` | Encoded waveform data (e.g., for display). |
| `remix_parent_id` | ID of the parent if this is a remix. |
| `duration_s` | Duration of the audio in seconds. |
| `privacy` | Privacy status (`public`, `private`, etc.). |
| `transform` | Any transformation applied (e.g., pitch shift, tempo change). |
| `parent_riff_id` | ID of the original riff if derived. |
| `audio_upload_id` | ID for uploaded audio, if any. |
| `model_display_name` | Name of the model used (e.g., `suno-v2`). |
| `is_favorite` | Boolean flag indicating if user favorited it. |
| `favorite_count` | Number of times favorited by others. |
| `play_count` | Total play count. |
| `allow_public_use` | Whether others can remix/use it publicly. |
| `can_use` | Whether the user has permission to reuse. |
| `audio_b64` | Base64-encoded audio (if stored inline). |
| `image_b64` | Base64-encoded image (if stored inline). |
| `video_id` | ID of generated video, if available. |
| `seed` | Random seed used for generation. |
| `conditions` | Input prompt or control data used for generation. |
| `lyrics_timestamped` | Timestamped lyrics (if aligned with audio). |
| `audio_url` | URL to the audio file. |
| `image_url` | URL to the generated image. |
| `video_url` | URL to the generated video. |

本次介绍Riffusion：
Riffusion是目前市场上颇具影响力的商用人工智能歌曲生成工具之一，同时也是早期高品质入局者之一。其正式发布之际便被盛赞为音乐领域的Stable Diffusion（稳定扩散），且曾被应用于Mila实验室推出的首批知名**虚假歌曲基准测试集**中。

本次我们将发布99293首歌曲及其配套的精细化元数据，以助力该领域的前沿研究。同时我们恳请所有使用者遵守本数据集的使用条款，具体如下：
1. 本数据集采用CC-BY-NC-ND许可协议，禁止对其进行商业使用。
2. 同时禁止将本数据集的任何衍生作品重新上传至互联网任何平台，亦包括本数据集本身的分享行为。

请遵守上述规则，以协助我们与所有研究者合法发布高品质的研究数据集。

> 特别鸣谢Azeem为本数据集的构建所付出的努力。

## 元数据字段说明

| 字段名 | 描述 |
|----------------------|----------------------------------------------------------------------|
| `id` | 生成内容的唯一标识符。 |
| `group_id` | 关联相关变体或混音作品的组ID。 |
| `author_id` | 创建该内容的用户或系统ID。 |
| `created_at` | 内容生成的时间戳。 |
| `audio_variation` | 音频变体的类型或版本（例如：原版、混音版）。 |
| `lyrics` | 与音频关联的原始歌词（未带时间戳）。 |
| `title` | 生成作品的标题。 |
| `image_override` | 指示是否使用自定义图像的标记或ID。 |
| `sound` | 音频风格的标签或分类。 |
| `topic` | 生成任务关联的主题。 |
| `image_id` | 生成或附加图像的ID。 |
| `simple_waveform` | 编码后的波形数据（例如用于展示）。 |
| `remix_parent_id` | 若为混音作品，则为其父作品的ID。 |
| `duration_s` | 音频时长，单位为秒。 |
| `privacy` | 隐私状态（如`public`（公开）、`private`（私有）等）。 |
| `transform` | 应用的任何变换操作（例如：音调偏移、速度调整）。 |
| `parent_riff_id` | 若为衍生作品，则为原始riff的ID。 |
| `audio_upload_id` | 上传音频的ID（若存在）。 |
| `model_display_name` | 所用模型的名称（例如：`suno-v2`）。 |
| `is_favorite` | 指示用户是否将其标记为收藏的布尔标记。 |
| `favorite_count` | 其他用户的收藏次数。 |
| `play_count` | 总播放次数。 |
| `allow_public_use` | 是否允许其他用户对其进行公开混音或使用。 |
| `can_use` | 用户是否拥有重复使用的权限。 |
| `audio_b64` | 内联存储的Base64编码音频数据（若存在）。 |
| `image_b64` | 内联存储的Base64编码图像数据（若存在）。 |
| `video_id` | 生成视频的ID（若可用）。 |
| `seed` | 生成所用的随机种子。 |
| `conditions` | 生成所用的输入提示词或控制数据。 |
| `lyrics_timestamped` | 与音频对齐的带时间戳歌词（若存在）。 |
| `audio_url` | 音频文件的URL。 |
| `image_url` | 生成图像的URL。 |
| `video_url` | 生成视频的URL。 |