qparse framework
The qparse library implements an original approach for the transcription of symbolic music performances into structured music scores (see the page on activities of the project)
The C++ sources (77k loc) are available in a public gitlab repository, see also the following page for a description of the tools.
https://qparse.gitlabpages.inria.fr
The project provides the following transcription tools
- monoparse, a command line tool for the transcription of monophonic melodies,
- drumparse, a command line tool for the transcription of recordings of electronic drumkits,
- pyqparse, a binding for running transcription from a Python interpreter, which is integrated with the toolkit Music 21 for computational musicology. Some Pythons scripts and use cases of this binding can be found in the following repository.
https://gitlab.inria.fr/lyrodrig/pyqparse
The Score Model presented in the page activities is part of the qparse library, due to mutual dependencies: qparse is used to construct a Score Model from unstructured music content, by parsing, and the Score Model export functionalities are used to provide scores in XML formats. This model is used in other activities and developments besides our transcription tools.
Music score diff
A first version of a diff tool, identifying differences between music scores, has been implemented in Python by Francesco Foscarin on the top of the data structures of Music 21 toolkit. It relies on original Dynamic Programming algorithms for computed an ad-hoc edit-distances on sequences of trees.
A second version is in development on the top of our (C++) Score Model, for improved performances and better integration with our other tools.
Another version is developed independently in the following fork:
https://github.com/gregchapman-dev/musicdiff
Pitch Spelling
The Pitch Spelling procedure mentioned in the page activities has been implemented in C++, with a Python binding and Python scripts for experimentation.
https://gitlab.inria.fr/pse/pse
An evaluation is ongoing in the following repository.
https://gitlab.inria.fr/pse/pseval
Neuma
The digital library of music scores Neuma has been developed by a consortium involving CNAM and IReMus, long before the Codex project. Members of Codex are contributing to this library. One can mention in particular the development of a RESTful service for melodic search.
Generation of performances
Our partners at Sakai Lab, Nagoya University, and Tojo Lab, JAIST Kanazawa are developing a toolkit for the generation of realistic music performances from XML scores, relying on segmentation into phrases, provided by automated musicological analyses performed in Tojo Lab, following the GTTM (Generative Theory of Tonal Music).
https://git.trs.css.i.nagoya-u.ac.jp/transcription/dm-env
Such performances provide and are in particular limitless and various amounts of data for the evaluation of transcription.
Datasets
See the page on activities of the project for a description of the context of the creation of the following datasets, and their use in the Codex project.
- The dataset Lamarque-Goudard, contains about 300 MusicXML scores of monophonic music, with several MIDI performances for each. The performances are generated automatically, or, more rarely, recorded by professional pianists on an electronic keyboard.
It is called by the names of the authors of a method for teaching rhythm in french music schools. - The dataset ASAP of Western classical piano music. contains 222 digital musical scores aligned with 1068 performances.
- The dataset scored GMD, containing scores transcription of the MIDI recordings in the Groove Midi Dataset, is currently under development.
