Our goal was to realise a computer-aided music learing system, called ``Virtual Music Teacher'', which would allow to detect errors in a played music piece and give explainations on these errors.
All error kinds have been described (2.3.1), input, output and other data to use have been defined (2.2 and 2.4). The system is based on score following using Hidden Markov Models (1.2).
The Virtual Music Teacher has been implemented as a package for jMax, which is a real-time programming environment for music and multimedia produced at IRCAM - Centre Pompidou, Paris.
Implementation work realised between march and august 2003 has led to a 1.0 version of the VMT package (VMT-v1.0) which has following features:
- Error detection can be made on a monophonic piece,
- MIDI data are used as input for the performance,
- Reference score is acquired with a text file, following a defined syntax,
- Detected errors are note errors (wrong, skip and added note), rhythm errors (note too long or too short, skip or added rests), and dynamic errors (note too loud or too soft, accentuation),
- An online signalisation is possible when an error occurs,
- It is possible to edit a report at the end of the play. This report contains : playing conditions, number of played errors, error details (location, kind, explaination).
VMT-v1.0 has been tested and gave its first results (ch. 6). Work to do and future work are described in next section.