Aria MIDI Collection of Solo Piano Recordings

Aria-MIDI contains 1,186,253 MIDI files representing approximately 100,629 hours of transcribed solo-piano music. The dataset was created by author loubb and includes metadata categories such as genre, composer, performer, and compositional identifiers. It was last updated on December 14, 2025.

Use Cases

• Pre-train generative music models using the 1.2 million MIDI files and their associated genre metadata.
• Analyze compositional patterns by studying the composer and performer metadata across the collection.
• Train classifiers to identify musical genres based on the transcribed MIDI data and its labels.
• Benchmark music information retrieval algorithms on the large-scale collection of solo-piano recordings.

Strengths

• Contains 1,186,253 individual MIDI files.
• Represents approximately 100,629 hours of audio content.
• Includes metadata categories like genre, composer, and performer.

Limitations

• Dataset scope is limited exclusively to solo-piano recordings, introducing a strong instrumental bias.
• Specific details on data collection method, geographic origin, and temporal coverage are unavailable.

Provenance

Source

loubb on Hugging Face

Collection Method

Collection of transcribed MIDI files; specific transcription method unknown.

Freshness

Last updated December 14, 2025.