Les, For some reason I missed this, likely this is because of my current flu and light fever... Oops. That looks very interesting indeed! You'll have to excuse me for being a little sceptical about applying neural networks to arbitrary problems because they are notoriously hard to optimise (in width, depth and method of interconnection, etc) and defining the feedback loop for machine-learning is non-trivial... But this looks like a fine issue to approach in a very naive and outgoing way. I should probably explain here that a long time ago (a decade or so) I used to study artificial intelligence, at least until I dropped out. Back then neural networks seemed like something magical in the future (of our studies), then again, so seemed OOP and ChucK made that look easy as well so I say; "go!". Yours, Kas. 2008/5/26 :

Hi Kassen,

I read about it in this previous message from this mailing list, see item 2. Maybe I'm confused. I'm easily amused and easily confused! Haha!

Les

Hi Nuno,

This would be a great application for our new SMIRK toolkit (small music information retrieval toolkit for MIR in ChucK), soon to be up at http://smirk.cs.princeton.edu.

This is a great example of a problem that could be easily solved with a machine learning algorithm, wherein you 1) Extract features from a training set of animal sounds 2) Use them to train a classifier (now available in chuck: kNN, adaboost) 3) extract features from the mic input 4) use the trained classifier to classify the new inputs

You could start with playing with FFT, centroid, RMS, rolloff, and other standard features, then use whatever features end up capturing your idea of similarity the best.

I'll send a notice to this list once everything is totally up.

Cheers, Rebecca _______________________________________________ chuck-users mailing list chuck-users@lists.cs.princeton.edu https://lists.cs.princeton.edu/mailman/listinfo/chuck-users