Interactive Music Transcription based on Rhythm Tree Languages

International audienceWe consider symbolic rhythm transcription, also called quantization: the problem of the conversion of a sequence of timestamped notes, e.g. a MIDI file, into conventional Western music notation. The fitness of a notation to a given a MIDI input is an intrinsically ambiguous notion, as it depends strongly on the the musical context.We propose a system available as a graphical component of the assisted composition environment OpenMusic, which privileges user interactions in order to search for an appropriate balance between different criteria, in particular the precision of the transcription and the readability of the musical scores produced.For the development of this system, we followed a uniform approach based on a set of preferred rhythm notations, specified a priori as a formal weighted tree language. Using efficient dynamic programming algorithms, we propose a on-the-fly enumeration of ranked solutions of quantization in this language, allow the user to explore and visualize the solution space, select rhythms and edit them