SingMOS-Pro: A Benchmark for Singing Voice Quality Assessment

SingMOS-Pro is an official benchmark dataset for singing voice quality assessment. The dataset was created by TangRain and released on Hugging Face, with a last recorded update on October 9, 2025. It is described as the official benchmark for the VoiceMOS 2024 singing track.

Use Cases

• Training models for singing voice quality assessment based on the benchmark's annotated samples.
• Benchmarking new audio quality models against the official VoiceMOS 2024 singing track.
• Researching the correlation between subjective and objective metrics for singing voice quality.

Strengths

• Serves as the official benchmark for the VoiceMOS 2024 singing track, providing a standard for comparison.
• Has an associated paper describing the dataset design and annotation methodology.
• Last updated on 2025-10-09, suggesting recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

TangRain on Hugging Face

Collection Method

Likely contains annotated singing voice samples, as described in the associated paper.

Freshness

Last updated 2025-10-09 06:47:20; freshness should be verified.