MMGenre: Chinese Singing Voice and Score Pairs Across 10 Genres

MMGenre is a multi-genre benchmark for singing voice synthesis containing 3,152 aligned Chinese singing voice and symbolic music score segments. The data, derived from 148 songs, totals approximately 261.8 minutes (4.36 hours) of audio. It was created by author Leaky-ReLU and last updated on the Hugging Face platform in June 2026.

Use Cases

• Benchmarking genre generalization in singing voice synthesis models based on the multi-genre structure.
• Training models for audio-score alignment using the provided paired segments.
• Analyzing acoustic characteristics across 10 major and 26 subgenres of Chinese music.
• Developing genre-conditioned or genre-transfer singing voice synthesis systems.

Strengths

• Contains 3,152 audio-score segments, providing a substantial number of data points for model training and evaluation.
• Covers a wide range of 10 major musical genres and 26 subgenres, enabling genre-specific analysis.
• Total audio duration is approximately 261.8 minutes (4.36 hours), offering a significant amount of training material.
• Segments are aligned pairs of singing voice and symbolic music scores, which is a structured format for synthesis tasks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific modeling needs.
• Data may reflect geographic or musical bias inherent to the source songs, focusing exclusively on Chinese music.

Provenance

Source

Hugging Face dataset uploaded by author Leaky-ReLU.

Collection Method

Derived from 148 songs, with audio and symbolic score pairs aligned and segmented.

Freshness

Last updated 2026-06-15 10:02:38; freshness should be verified.

Geography

The singing voice data is in Chinese, suggesting a potential geographic focus on Chinese-language music.