Stability AI has launched a new artificial intelligence model designed for text-to-audio generation, developed in collaboration with Arm. The model, named Stable Audio Open Small, was unveiled on Wednesday and is capable of producing short audio samples from text prompts. According to the London-based firm, this model is lightweight, optimized to operate entirely on Arm CPUs, and features a rapid audio generation time, making it well-suited for bulk applications. Developers can access the open-source audio model for download on both GitHub and Hugging Face.
Stability AI Releases Stable Audio Open Small
In a recent announcement posted on their news platform, Stability AI outlined the features of this new large language model. Stable Audio Open Small is a distilled version of the Stable Audio Open model, which debuted in June 2024, and it is capable of generating audio up to 47 seconds in length. The redesign prioritizes faster generation and a more compact size.
This new model boasts 341 million parameters and can produce audio samples of up to 11 seconds. The company asserts it can deliver an audio sample in under eight seconds while running locally on smartphones. Notably, the collaboration between Stability AI and Arm for this generative audio project was first revealed at the Mobile World Congress (MWC) 2025.
Regarding its underlying structure and training, Stable Audio Open Small utilizes a latent diffusion model built on a transformer architecture. It has been trained on an extensive dataset composed of 486,492 licensed audio recordings. For text conditioning, the model leverages a publicly available pre-trained T5 model. Additionally, the post-training phase employed an Adversarial Relativistic-Contrastive (ARC) algorithm to enhance prompt adherence and boost inference speed.
The company indicates that this text-to-audio model is well-suited for generating drum loops, foley effects, instrument riffs, and ambient sound textures. Its compact size enables deployment on Arm-powered smartphones and various edge devices, making it ideal for applications requiring real-time generation and high responsiveness.
Model weights for Stable Audio Open Small can be downloaded from the AI firm’s Hugging Face listing, while the associated code base is available on GitHub listing. The AI model is accessible for both commercial and non-commercial purposes under the Stability AI Community License.
