At this year’s ISMIR (International Society for Music Information Retrieval) online conference we introduced a workflow paradigm for remote consensus building between expert annotators, and showcased one possible implementation by releasing a (nearly completed) Corelli dataset created with it. It includes the scores of 36 sonatas (opp. 1, 3, and 4) in MuseScore 3 format that include expert annotations of harmony, phrases, and cadences.
The conference paper focuses on the new workflow paradigm which applies the idea of distributed version control ubiquitous in software development (e.g. in the form of Git) in order to create fully validated and verified labels while automating tedious and repetitive tasks. The DCML’s implementation uses GitHub Actions to trigger Python scripts upon particular events (red boxes in the figure):
For a small run through you can watch this 4-minute pitch:
The code of our GitHub implementation of the paradigm is open to the public and can be re-used by referring to this manual. For more detailed information, please refer to the paper: