Stability AI open-sources an audio model that can run on smartphones with Arm CPUs

Stability AI has released Stable Audio Open Small as an open-source text-to-audio model featuring 341 million parameters and optimization for Arm central processing units. Developed in cooperation with Arm, the model targets on-device deployment, allowing audio generation directly on mobile hardware.

Stable Audio Open Small is engineered to rapidly generate brief audio samples. Users can achieve up to 11 seconds of audio in under 8 seconds when running the model on a smartphone equipped with an Arm CPU. This efficiency makes the model accessible for mobile app developers and those interested in generative sound applications without relying on remote servers.

Following prior collaboration with Arm, this release aims to expand generative audio capabilities on smartphones. Developers interested in practical implementation can consult the new Arm Learning Path, which provides step-by-step, hands-on guidance for deploying the model on compatible hardware.

The model is available under a permissive Stability AI Community License, covering both commercial and non-commercial usage. Additional resources are also provided, including a research paper on arXiv, model weights on Hugging Face, and source code on GitHub.