This page gives sample results of an algorithm that uses genetic algorithms to evolve waveforms. For a cursory introduction to genetic algorithms and a brief overview of the algorithm, please see the overview page. For a more info, please see my research page.

Results

All subsequent links on this page are to wave files, some of which are quite large.

Test Sets:

• Population of sinusoids evolving in the environment of a sinusoidal glissando

  

• Population of white noise evolving in the environment of a sinusoidal glissando

  

• Population of samples taken arbitrarily from a composition for percussion and voices evolving in the environment of the single ring of a bell.

Without Mutation

• Sexual reproduction of sinusoids without mutation.

  

With Mutation

Mutation by Amplification:

• Typical results of long-term evolution of sinusoids.

  

• Typical results of long-term evolution of noise.

  

• Typical results of long-term evolution of real-world samples.

Mutation by Exponentiation:

• Typical results of long-term evolution of sinusoids.

  

• Typical results of long-term evolution of noise.
• Typical results of long-term evolution of real-world samples.

Mutation by Swapping, Reversal, or Duplication:

• The typical results of long-term evolution of sinusoids is similar for these three cases.

  

Mutation by Reversal:

• The top waveform is the original. The bottom waveform shows typical results of long-term evolution of real-world samples.

  

Mutation by Swapping:

• The top waveform is the original. The bottom waveform shows typical results of long-term evolution.

  

• Typical results of long-term evolution of noise.
• Typical results of long-term evolution of real-world samples.

Mutation by Duplication/Elimination:

• The top waveform is the original. The bottom waveform shows typical results of long-term evolution evolution of noise. This sample is longer than the others and makes clear the algorithm's potential for constantly changing results of similar character over time.

  

• Two examples of real-world samples evolving. The first abstracts out the pitch of the target environment; the second abstracts out the envelope and some features of the pitch. This demonstrates the ability of the algorithm to find divergent results from the same starting conditions.