I think it’s about time I introduce my project:

I want to use Pd to extract the overtones of the recorder and train a neural network to identify what kind of sound (broad, thin, glass-ish, etc) I’m playing. I want to use this to control VST processors to shape the sound I play. And after the network is trained, I want to the analysis and processing in real-time.

At the moment I’m reading up on alternatives to FFT to separate out overtones, and I’ve made a prototype as a proof-of-concept kind of thing using fiddle~ to identify the overtones in hz and bp~ to separate them out as audio streams. It still has lots of problems such as artifacts, sound degeneration and latency issues. But it gets the job done, so I’m optimistic.

I’d prefer to do my project as a part of a Ph.D, but since Ph.D funding takes a while to find, I’m already starting. Can’t wait for formalities. :-) When I’ve checked that it is ok to release code here I’ll post some as I progress.

Categories: Technology