Beethoven Symphony Excerpt Dataset (BSED): An Evaluation Dataset for Orchestral Music Transcription
Authors/Creators
Description
The Beethoven Symphony Excerpt Dataset (BSED) is a dataset primarily designed to support the evaluation of Automatic Music Transcription (AMT) for orchestral music. The accompanying Beethoven Symphony Dataset (BSD) serves as a larger training dataset. While primarily developed for AMT, both datasets can also be used for a range of Music Information Retrieval (MIR) tasks involving orchestral music, such as score-audio alignment and music synthesis.
BSED
The BSED comprises 20 score excerpts from Beethoven's symphonies. For each excerpt, we provide five synchronized audio versions: four real concert recordings and one synthetic rendition. The corresponding symbolic scores are available in MusicXML, Sibelius, MIDI, and CSV note-event formats. We obtain note-level annotations for each audio recording by aligning the symbolic scores to the audio using a robust score-audio alignment pipeline, followed by manual verification and refinement to ensure high temporal accuracy.
BSD
The BSD comprises public-domain audio recordings of Beethoven's complete symphonies, totaling approximately 62 hours of audio material. As with the BSED, note-level annotations are obtained by aligning symbolic scores to the audio recordings. However, due to the considerably larger scale of the dataset, the alignments have not undergone the same level of manual verification as those in the BSED. Consequently, the BSD serves as a valuable large-scale training resource, offering substantial diversity in performances while providing slightly less rigorously validated annotations than the BSED.
For more information and tools to recreate and interact with the data, please have a look at the accompanying GitHub repository. [link to follow]
If you find any errors or problems with the dataset, please raise an issue in the GitHub repository.
Please cite the following paper if you use our work:
Berendes, H.-U., Saha, A., Maman, B., Arifi-Müller, V., & Müller, M. (2026). Beethoven Symphony Excerpt Dataset (BSED): An Evaluation Dataset for Orchestral Music Transcription. Transactions of the International Society for Music Information Retrieval, 9(1), 405–422. DOI: https://doi.org/10.5334/tismir.343
Files
BSD.zip
Additional details
Funding
- Deutsche Forschungsgemeinschaft
- Learning with Music Signals: Technology Meets Education No. 500643750
Software
- Repository URL
- https://github.com/groupmm/bsd-code