We introduce a novel method for generating arbitrary additive synthesizers for audio replication: given a target audio signal, digital signal processes, expressed as cyclic graph structures, are evolved via Recurrent Cartesian Genetic Programming (RCGP) to generate an approximation of the target audio with arbitrary accuracy. The Log-Spectral Distance (LSD) between the target and actual output signals is used to measure the fitness of each solution candidate. The candidate with the smallest LSD is returned as the solution. A working implementation of RCGP in C that generates Functional AUdio STream (FAUST) code is provided, then evaluated via re-synthesis of steady state tones as provided in the Sandell Harmonic ARChive (SHARC). Early results show promise as a mean LSD value of 5.1 dB was achieved after 1000 generations, with 0.52 dB being the minimum LSD found.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=21954
(228KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
