An AI that turns text into a synth sound
- Context
- Epitech final-year project, team of three. I handled the AI side and the backend.
- Year
- 2025–2026
- Stack
- Python
- BERT-tiny
- Flask
- C++
- JUCE
- Links
An AI engine that turns a written description (“a warm, slow pad”) into synthesizer settings.
My part
- Build the dataset that pairs descriptions with synth settings.
- Fine-tune a small open-source model (BERT-tiny) on that dataset.
- Build the pipeline linking the two models.
- Develop the application backend and the Flask API that serves inference to the plugin.
Architecture
- 01Prompttext description
- 02BERT-tinyfine-tuned model
- 03Flaskinference API
- 04VSTC++ audio plugin (JUCE)