Syntheon

Syntheon - Pantheon for music synthesizers.

Syntheon provides parameter inference for music synthesizers using deep learning models. Given an audio sample, Syntheon infers the best parameter preset for a given synthesizer that can recreate the audio sample.

**Check out this presentation on the recent advances of synth parameter inference.

For now:

• :heavy_check_mark: Vital is supported
• Dexed is work-in-progress

Try it out on our Colab notebook demo.

Installation

python3 -m pip install syntheon

Usage

from syntheon import infer_params

output_params_file, eval_dict = infer_params(
    "your_audio.wav", 
    "vital", 
    enable_eval=True
)

Testing

python3 -m pytest

Structure

For each synthesizer, we need to define:

• converter for preset format conversion:

  • serializeToDict: convert preset file to a Python dictionary to be handled by inferencer
  • parseToPluginFile: convert Python dictionary back to preset file, to be loaded by the synthesizer

• inferencer for model inference:

  • convert: define the workflow of load_model -> inference -> convert_to_preset

Contribution

Syntheon is actively under development, and contributions are welcomed. Some TODOs we have in mind include:

• Replicating state-of-the-art approaches
• Improving current model performance
• Incorporating new synthesizers
• Code refactoring 😅