Stability AI, Arm Release Lightweight Text-to-Audio Model for Smartphones

Stability AI, in partnership with Arm, has released Stable Audio Open Small, an open-source, lightweight text-to-audio AI model optimized to run entirely on Arm CPUs, enabling fast, offline audio generation on smartphones and edge devices. The model, with 341 million parameters, can produce up to 11 seconds of stereo audio in under eight seconds, making it the fastest of its kind for on-device use. Unlike many competitors that rely on cloud processing, Stable Audio Open Small operates efficiently on-device, enhancing accessibility while lowering compute costs. Stability AI highlights its ethical training approach, using only royalty-free audio from Free Music Archive and Freesound, avoiding intellectual property issues faced by others. However, the model currently supports only English prompts, cannot generate realistic vocals or high-quality songs, and its performance varies across musical genres due to Western-centric training data. The model is free for researchers, hobbyists, and small businesses, with enterprise licenses required for larger companies, and is available with full documentation, code, and weights on platforms like Hugging Face and GitHub.