Skip to content

Latest commit

 

History

History
39 lines (26 loc) · 1.22 KB

README.md

File metadata and controls

39 lines (26 loc) · 1.22 KB

logo

Generate synthesizer sounds from text prompts with a simple evolutionary algorithm.

Audio examples here: https://erl-j.github.io/textsynth/

Synth: https://github.com/torchsynth/torchsynth

Audio-Text cross modal embedding: https://github.com/LAION-AI/CLAP

How it works

Start with randomly initialized synthesizer sounds. Each iteration, the current synthesizer sounds are evaluated on how well they match the text prompt. The best sounds are then combined and mutated to generate new sounds for the next iteration. 200 generations w/ 50 samples takes about ~20s on a 3090 (not tested on CPU).

example

Future work

  • Install guide / requirements.txt
  • Diversity preservation.
  • Open ended exploration.
  • RL
  • Neural nets?

Cite

If you found this project useful please cite:

@software{
  textsynth,
  author       = {Nicolas Jonason},
  title        = {TextSynth: Generate synthesizer sounds from text prompts with a simple evolutionary algorithm},
  month        = october,
  year         = 2023,
}