(Nederlandse versie: ComParser)
ComParser is experimental (but free) software for realtime audio recognition and a very crude form of musical score following. The software is being developed and applied by the Dutch Schreck Ensemble.
The approach differs from for example IRCAM: hidden markov models and high-level representation, the symbolic approach.
ComParser works with artificial neural networks
and follows the more sonic approach
- ComParser is written in ISO/ANSI C (and it will stay that way).
- ComParser's realtime audio i/o is based on PortAudio.
- ComParser read and write audiofiles in the AIFF format.
Download software:
Trouble with the current ComParser (up to version 1.25):
- Limited dynamic range.
- Excessive spectral dependancy.
About the comparser~PD object:
On march 23, 2005, I did an inofficial ComParser-demo at the Music Technology Group at the UPF in Barcelona. Got a nice Python-script from Ali Momeni: http://cnmat.cnmat.berkeley.edu/%7eali/share/max/scripts/pd-to-max.sit to convert PD-objects to Max-objects. (thanks!)
However, it may perhaps be better to use the PD/MAX-abstraction layer: http://grrrr.org/ext/flext/ since the sourcecode may keep changing.
(-Python is an interesting language when applied to structural needs of composers and
constructivists in general. Flext's author is Thomas Grill, who is also an avid
SuperCollider user... /There's is a striking similarity between certain parts of
the Python and SC language. Probably a Smalltalk-aspect.../-)
Another nice suggestion from Ali is to seperate the audio-preprocessing from the neural network.
[-Because it opens up the the access to the the network with additional analysis-techniques,
thus refining it.(as I suggested previously in private conversations-AvS)]
Indeed!: that is the way to go.