Stability AI and Arm drop an audio model that runs entirely on your phone
Stability AI just teamed up with Arm to open-source Stable Audio Open Small, a 341M parameter text-to-audio model built to run on smartphone CPUs. Powered by Arm KleidiAI, it generates up to 11 seconds of stereo sound in under 8 seconds. Why it matters: generative audio is finally moving off the cloud and directly onto edge devices for real-time sound effects.
Stable Audio Open Small is a lightweight text-to-audio model from Stability AI and Arm, built to generate stereo sound effects directly on a smartphone CPU without touching the cloud.
What it does
This model is designed for quick generation of short audio samples and production sound design. By typing in a simple text prompt, you can create drum loops, foley, instrument riffs, and ambient background textures.
It produces up to 11 seconds of stereo audio on demand. Stability AI claims it is currently the fastest stereo text-to-audio model available.
The previous Stable Audio Open model packed 1.1 billion parameters, but this new Small variant trims that footprint down to 341 million parameters. Despite the drastic diet, Stability AI says it keeps output fidelity and prompt adherence intact.
How it runs
The clever move here is skipping mobile GPUs and dedicated NPUs entirely. The model is specifically tuned to run on standard Arm CPUs, powered by Arm's KleidiAI libraries.
On an Arm-powered smartphone, generating an 11-second audio clip takes under 8 seconds. Because the footprint is so light, it is also significantly faster and cheaper to fine-tune on custom audio datasets.
Stability AI open-sourced the model under its Community License, making it free for both commercial and non-commercial use. The model weights are live on Hugging Face, code is on GitHub, and the technical paper is available on arXiv. Developers looking to build can follow the Arm Learning Path guide for hands-on deployment.
Why it matters
For mobile developers and audio toolmakers, this eliminates the two biggest headaches of AI audio: latency and cloud API costs. Being able to spit out bespoke sound effects locally on a phone makes interactive sound design possible even in offline mode.
It also sets a smart template for edge deployment. Full-length music tracks still make sense in the cloud, but allocating short, repetitive foley tasks to mobile CPUs is a practical way to balance compute budgets.
Sources (1)
- stability.ai(opens in a new tab)Originalstability.ai