OpenScore Lieder: 19th-Century Song Scores for Optical Music Recognition Evaluation

OpenScore Lieder is a dataset derived from a corpus of 19th-century songs in MuseScore/MusicXML format, created by Gotham & Jonas in 2022. It is built for evaluating Optical Music Recognition systems under piano-only, full-page conditions. The dataset provides both camera page images from source PDFs and clean page images rendered from ground-truth MusicXML.

Use Cases

• Benchmarking Optical Music Recognition (OMR) system accuracy based on piano-only, full-page conditions.
• Training OMR models using paired camera page images and clean rendered images.
• Researching OMR performance on 19th-century musical notation from the Lieder corpus.

Strengths

• Derived from a published corpus (Gotham & Jonas, 2022) providing a known source.
• Includes both camera page images and clean rendered images for evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Data may reflect temporal bias inherent to the 19th-century source material.

Provenance

Source

OpenScore Lieder corpus (Gotham & Jonas, 2022).

Collection Method

Source PDFs used to obtain camera page images; clean page images rendered from ground-truth MusicXML.

Time Range

19th-century.

Freshness

Last updated 2026-04-03 23:03:06; freshness should be verified.

Geography

null