Erl-j's Soundfont Generator

Generate soundfonts from text descriptions using latent flow matching. You can then download the complete SFZ soundfont package to use the instrument locally.

Instructions

  1. Enter a text prompt to describe the audio you want to generate. For example, "hard bass" or "dark analog lead flute flute flute soft".

  2. Adjust the number of generation steps to tradeoff between quality and speed (kindof).

  3. Click the "Generate Soundfont" button to generate the audio and soundfont.

  4. Preview the generated instrument with the keyboard.

  5. Export the soundfont by clicking the "Download SFZ Soundfont Package" button. You can then use the soundfont in a SFZ-compatible sampler. See this list for a list of SFZ players.
1 50

Download Soundfont Package here:

About

The model is a modified version of stable audio open.

Unlike the original model, this version uses latent flow matching rather than latent diffusion. Secondly, the pitches are stacked in a channel dimension rather than concatenated in the time dimension. This allows for faster generation.

Soundfont export code is based on the sf-creator project.

Similar work by Nercessian and Imort: InstrumentGen.

Thank you @carlthome for coming up with the name.

To cite this work, please use the following BibTeX entry:

@misc{erl-j-soundfont-generator,
    title={Erl-j's Soundfont Generator},
    author={Nicolas Jonason},
    year={2024},
    publisher={Huggingface},
}