Beautiful Motifs: 319k+ High-Rated Music Motifs Extracted from MIDI Files

319,000+ high-rated beautiful music motifs extracted from 248,000+ high-quality MIDI files from the Discover MIDI dataset. The dataset was created by asigalov61 using a beauty metric based on cosine similarity of embeddings against a reference set, with scores ranging from 0.739 to 0.999. It was last updated on June 19, 2026.

Use Cases

• Training generative music models based on high-quality motif embeddings.
• Analyzing musical beauty or similarity based on the provided cosine similarity scores.
• Creating datasets for music recommendation or classification tasks using the extracted motifs.
• Studying patterns in high-quality MIDI compositions from the referenced source dataset.

Strengths

• Contains over 319,000 music motifs, providing a substantial collection.
• Derived from a curated set of over 248,000 high-quality MIDI files.
• Includes a quantitative beauty score for each motif, with a mean score of 0.9057.

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

huggingface dataset by author asigalov61, referencing the Discover MIDI and Beautiful Music Seeds datasets.

Collection Method

Motifs were extracted and rated using the 'midisim' method and a beauty metric based on top k embeddings cosine similarity against a transposed reference MIDI dataset.

Freshness

Last updated 2026-06-19 04:13:00; freshness should be verified.