Paper at AES 130
My friend and collaborator Matthew Yee-King presented our joint paper, “A Comparison of Parametric Optimisation Techniques for Tone Matching” at the Audio Engineering Society conference 130 on Sunday, May 15 in London. We started work on the topic beginning in October 2007 under the project name “Synthbot”.
The original goal of the project was to produce software capable of finding the presets for any synthesizer (i.e. VST plugin) to produce any given sound (or as close as possible). This spawned a number of research problems, including psychoacoustic sound similarity, determination of error surfaces of musical synthesizers, optimisation techniques for such error surfaces (featuring discrete and continuous parameters), and characterisation and modelling of synthesizer timbre-space, among others. I worked on this problem for about two years in my spare time (as it was far more interesting than writing instant messages which was my day job at the time). We were able to achieve good results with FM and subtractive synthesizers with a relatively low number of parameters (dozens). Larger synths (i.e. hundreds of parameters) proved to have too large a parameter space to search effectively with our techniques (the 1000+ parameter FM8 was absolutely untouchable). Data driven approaches worked much better at that size. Using what we learned, we were also able to automatically generate presets for any given synth which effectively covered the types of sounds possible by the synth.
During the conference we were lucky enough to meet the makers of SoundTorch who were working on exactly the same problem. This was very serendipitous and lead to interesting discussions, comparing our respective approaches (which were more or less the same).
