GROWING A SOUND SYNTHESIZER
Creation of sound synthesis algorithms using evolutionary methods
| By: Ricardo A. Garcia | rago@ragomusic.com |
In a nutshell:
Approach:
The proposed approach represents the Sound Synthesis Algorithms (SSA) as
topologies, and uses evolutionary methods (i.e. Genetic Programming) to evolve
populations of "candidate solutions".
(tree representation of a Sound Synthesis Algorithm)
(topology graph representation of a Sound Synthesis Algorithm)
Genetic Programming:
A population of topologies (individuals) is evaluated every generation, and a
fitness value is assigned to each one of the individuals. A new population is
created by probabilistic selection of the best-fitted topologies from the last
generation. Genetic operations are then performed on them to create new
"improved" individuals.
In this approach, each individual is modified in two stages:
1. The actual topology layout is changed (by using genetic operations)
2. Its internal parameters are optimized for the given target sound
Uses:
If you have a SSA that reproduces a particular sound, several things can be
done:
Example: evolution of a SSA for a piano note (C3)
In the figure sequence it can be seen the waveform, magnitude spectrogram and
topology for the best individual of some selected generations. Note that the
topologies vary drastically until the solution converges to a steady
configuration. The fitness function applied measured the analytical distance of
the magnitude and phase spectrograms to those of the Target sound.
|
|||
|
|||
|
|||
|
|||
|
|||
|
|||
|
