Muse: 116,000 Synthetic Songs with Lyrics and Style Descriptions

Muse contains 116,000 synthetic music tracks in Chinese and English, synthesized using SunoV5 and paired with automatically generated lyrics and style descriptions. Created by bolshyC and introduced in early 2026, the collection supports research into reproducible long-form song generation. The data is divided into Chinese (CN) and English (EN) subsets to facilitate multilingual audio modeling.

Use Cases

• Training text-to-audio models using the style descriptions and audio pairs
• Generating long-form music sequences based on the provided lyrics
• Benchmarking multilingual music synthesis across the CN and EN subsets

Strengths

• 116,000 synthetic song records
• Bilingual coverage in Chinese and English
• Includes paired lyrics and style descriptions for every track

Limitations

• Synthetic audio synthesized by SunoV5 may not capture human performance nuances
• Lyrics and style descriptions are automatically generated rather than human-annotated
• Lack of explicit column metadata in the repository

Provenance

Source

Muse: Towards Reproducible Long-Form Song Generation with Fine-Grained Style Control (Arxiv 2601.03973)

Collection Method

Synthetic generation using SunoV5 for audio and automated tools for lyrics and style descriptions

Freshness

Last updated January 2026.

Geography

Global (Chinese and English languages)