Reverse Engineering Audio Synthesiser Sounds

Harrison Taylor


Supervised by Dave Marshall; Moderated by Yukun Lai

The basic idea is given a set of sample recordings of a sound can you determine the appropriate parameters of a given synthesiser to best approximate the set of sounds.

This could be achieved by applying advance machine learning and, specifically, optimisation techniques.

The synthesiser (and all it's available parameters) would need to under computer control so this would ideally suit software based synthesisers but some midi enable hardware device could be used (if available).

In fact is VSTi/Audio Unit devices were used it may be possible to develop an test benches whereby the approximation of parameters could be developed for a wide variety of "soft synths".

Initial Plan (05/02/2018) [Zip Archive]

Final Report (11/05/2018) [Zip Archive]

Publication Form