TTxdfcgv7/suno

---
pretty_name: Suno Music Generation Dataset
size_categories:
- 100K<n<1M
task_categories:
- audio-classification
- text-to-audio
annotations_creators:
- found
language:
- en
- ja
- multilingual
license: cc0-1.0
multilinguality:
- multilingual
source_datasets:
- original
tags:
- audio
- video
- image
- text
---
# Dataset Card for Suno.ai Music Generation
### Dataset Summary
This dataset contains metadata for 659,788 songs generated by artificial intelligence on the [suno.com](https://suno.com) platform, a service that generates music using artificial intelligence. The songs were discovered by search queries with words from the [dwyl/english-words](https://github.com/dwyl/english-words) word list.

### Languages
The dataset is multilingual with English as the primary language:
- English (en): Primary language for metadata and most lyrics
- Japanese (ja): Present in some song lyrics and titles
- Other languages may appear in song lyrics and titles

## Dataset Structure

### Data Fields
The dataset is stored in Parquet format with zstd compression. Each record includes:

**Core Fields:**
- `id`: Unique identifier for the song (string)
- `video_url`: URL of the video version (string)
- `audio_url`: URL of the audio file (string)
- `image_url`: URL of the song thumbnail (string)
- `image_large_url`: URL of the large cover image (string)
- `is_video_pending`: Video processing status (boolean)
- `major_model_version`: Version of AI model used (string)
- `model_name`: Name of the model used (string)
- `is_liked`: Like status (boolean)
- `user_id`: Creator's ID (string)
- `display_name`: Creator's display name (string)
- `handle`: Creator's handle (string)
- `is_handle_updated`: Handle update status (boolean)
- `avatar_image_url`: Creator's avatar URL (string)
- `is_trashed`: Deletion status (boolean)
- `created_at`: Creation timestamp (string)
- `status`: Generation status (string)
- `title`: Song title (string)
- `play_count`: Number of plays (integer)
- `upvote_count`: Number of upvotes (integer)
- `is_public`: Public visibility status (boolean)
- `persona`: Creator persona information (JSON string, nullable)

**Metadata Fields (flattened from nested structure):**
- `metadata_tags`: Musical style and genre tags (string)
- `metadata_prompt`: Lyrics/prompt used to generate the song (string)
- `metadata_type`: Generation type (string)
- `metadata_duration`: Length of song in seconds (float)
- `metadata_refund_credits`: Refund status (boolean)
- `metadata_stream`: Streaming availability (boolean)
- `metadata_gpt_description_prompt`: GPT description prompt (string)
- `metadata_has_vocal`: Whether the song has vocals (boolean)
- `metadata_negative_tags`: Negative style tags (string)
- `metadata_error_message`: Error message if generation failed (string)
- `metadata_error_type`: Error type if generation failed (string)
- `metadata_artist_clip_id`: Artist clip reference ID (string)
- `metadata_cover_clip_id`: Cover clip reference ID (string)
- `metadata_edit_session_id`: Edit session ID (string)
- `metadata_stem_from_id`: Stem source ID (string)
- `metadata_persona_id`: Persona ID used for generation (string)
- `metadata_task`: Generation task type (string)
- `metadata_is_audio_upload_tos_accepted`: TOS acceptance status (boolean)
- `metadata_concat_history`: Concatenation history for extended songs (JSON string)
- `metadata_history`: Generation history (JSON string)
- `metadata_infill`: Infill generation parameters (JSON string)

### Data Splits
All 659,788 songs are in a single train split.

## Additional Information
### License
This dataset is dedicated to the public domain under the Creative Commons Zero (CC0) license. This means you can:
* Use it for any purpose, including commercial projects.
* Modify it however you like.
* Distribute it without asking permission.
No attribution is required, but it's always appreciated!
CC0 license: https://creativecommons.org/publicdomain/zero/1.0/deed.en