Abstract:

This paper describes a methodology for building a benchmark to evaluate automatic music transcription tools, i.e., softwares that can extract music information from digital audio signals. There are numerous applications to these tools, like data acquisition for music virtual libraries and computer-aided music learning for instance. The main purpose of this benchmark is to present a method to ranking the current tools, as well as the characterisation of their mistakes and the respective causes. We expect that the results will be valuable both to users, guiding their choice for the most suitable tool according to their needs, and to tools' developers, showing any essential improvements required.