Stability partners with Arm to bring its audio generation model for on-device mobile use

Stability AI has teamed up with chipmaker Arm to optimize its Stable Audio Open model for mobile devices running on Arm chips. This AI model can generate audio, such as sound effects, from text descriptions, and now it can do so offline on mobile devices without needing an internet connection.

The collaboration involved "distilling" the model to enhance performance, reducing audio generation time by 30 times. Now, an 11-second audio sample takes only about 8 seconds to generate on an Armv9 CPU, thanks to Arm's KleidiAI libraries that enable on-device processing.

While no detailed technical specifications or research papers have been released about these optimizations, Stability AI plans to extend this partnership to adapt other AI models for image, video, and 3D generation to mobile platforms. Although the Stable Audio Open model is not yet available for download, Stability aims to integrate it into consumer apps and devices in the future.