Timbre analysis

Today I finally had a chance to watch a decent display of the data outputted by my implementation of a critical band filter convolved with a amplitude spectrum generated by a fast fourier transform (FFT). The actual filter transformation I wrote almost two years ago, but due to other work getting in the way, it isn’t until now I have had a chance to implement a graphical display of the spectrum and actually test it. It works well with real time input and, as far as I can tell after a few short tests, it is relatively invariant to pitch and discriminates well between vowels. In other words; it behaves as I had hoped it would.

This feels like a major break through in the part of my project that deals with the more technical, hands on, aspect of timbre based interaction. It will be very interesting to patch in JetMap (an ANN implementation) and see what kind of results we might get from that. However, I will begin with making some overall adjustments and optimizations in the code and combine all the parts into one piece with a simple GUI to control all the parameters. Also, I should add a threshold to descriminate signal from noise (i.e. signal from no signal) and optimizations of the input vector (subtration of the average from all bins and perhaps normalization).

I will post some more data here within the next few weeks.