Stability AI, in collaboration with Arm, has announced the release of Stable Audio Open Small, a 341 million parameter text-to-audio model designed to run entirely on Arm CPUs. This open-source model, now available for both commercial and non-commercial use under the Stability AI Community License, enables developers to generate short audio clips,up to 11 seconds,on smartphones in under 8 seconds. The release builds on a recent demonstration at Mobile World Congress, where Stability AI showcased the model’s ability to create AI-generated audio on mobile devices.

The new model is a compact version of Stable Audio Open, which has 1.1 billion parameters. Stable Audio Open Small maintains high output quality and prompt accuracy while being significantly smaller and faster. Optimized with Arm’s KleidiAI libraries, it is tailored for efficient on-device performance, reducing compute costs and eliminating the need for heavy hardware. This makes it ideal for edge devices like smartphones, where speed and responsiveness are critical.

Stable Audio Open Small is designed for generating short audio samples, such as sound effects, drum loops, instrument riffs, and ambient textures. Its lightweight design and fast inference times make it well-suited for real-time applications on Arm-powered devices, which power 99% of smartphones globally. The model’s efficiency allows developers to align compute resources with specific tasks, such as creating brief audio elements rather than full-length tracks.

Developers can access the model weights on Hugging Face, explore the code on GitHub, and read the research paper on arXiv for detailed insights into its technical advancements. Additionally, Arm has launched a dedicated Learning Path to guide users through deploying the model on Arm hardware, complemented by a technical overview on the Arm Community Blog.