Researchers have introduced OSSQ-OMR, the first dataset specifically designed for optical music recognition (OMR) of multi-part musical scores, particularly string quartets. This dataset, derived from the OpenScore String Quartet corpus, pairs scanned music sheets from IMSLP with their digital transcriptions. It includes over 24,000 system images and 98,000 staff images, with transcriptions available in Extended Linearized MusicXML, kern, and ABC formats. A benchmark protocol and baseline results from LSTM and Transformer models are also provided, highlighting the impact of encoding choices and the models' performance on scanned versus synthetic inputs. AI
IMPACT This dataset could advance research in optical music recognition for complex, multi-part musical scores.
RANK_REASON The item describes a new dataset and benchmark for a specific research task (Optical Music Recognition), published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]
- Dasaem Jeong PhD
- Extended Linearized MusicXML
- International Music Score Library Project
- kern
- long short-term memory
- OpenScore String Quartet corpus
- OpenScore String Quartet for Optical Music Recognition
- Transformer++
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
