The AIRP: Finetuning AI with synth sounds
The AIRP (AI ARP) project explores the role of AI in music creation, and the role of participatory data collection in collaborative workflows. Using the duo recording session and other collaborative recordings on the ARP 2500, I trained a model to produce sounds similar to the instrument.
To fine-tune a generative model, I used Stable Audio Open, an open-source system chosen for its transparency and adaptability. A custom pipeline processed the recordings and annotations into a format suitable for training. Balancing technical parameters like sample rate and training precision, I iteratively refined the model through 800 training steps. The fine-tuned model was then hosted on a Gradio platform, creating an interface to generate ARP-style samples on demand.
The final track is constituted entirely of AI-generated samples, sliced and mixed in Ableton.
Generative audio