Stability AI Launches Lightweight Text-to-Audio Model

On May 14, 2025,Stability AI, in collaboration with Arm, has unveiled a new open-source artificial intelligence model called ‘Stable Audio Open Small’.

This compact, fast text-to-audio generation tool is designed to run entirely on Arm CPUs, including smartphones and edge devices.

What Is Stable Audio Open Small?

Stable Audio Open Small, is a lightweight version of Stability AI’s earlier Stable Audio Open model (released in June 2024).

While the original version can generate up to 47 seconds of audio, the new model is optimized for speed, portability, and on-device use.

It can generate up to 11 seconds of audio in under eight seconds, even on smartphones.

Key Features & Benefits

Compact Size: The model has 341 million parameters, making it ideal for low-power devices.

Speed & Efficiency: Designed for real-time audio generation, especially useful for creating drum loops, sound effects, instrument riffs, and ambient textures.

Offline Capability: Unlike many cloud-based tools, it runs locally, offering fast and private processing.

Open Source: Available on GitHub and Hugging Face, it can be used for both commercial and non-commercial purposes under the Stability AI Community License.

Technical Background

Architecture: Built using a latent diffusion model with transformer architecture.

Training Data: Trained on 486,492 licensed audio files, all royalty-free, addressing IP concerns.

Text Input: Uses a public pre-trained T5 model for understanding prompts.

Post-Training: Enhanced with the Adversarial Relativistic-Contrastive (ARC) algorithm to boost speed and prompt accuracy.

Limitations & Considerations

Stable Audio Open Small model currently supports only English text prompts and it not optimized for realistic vocals or full-length songs.

This model Training data has a Western bias, which may affect global musical variety.

Its free for researchers, hobbyists, and small businesses (under $1 million revenue). Enterprise license required for larger organizations.

Significance

Stability AI’s release marks a step toward democratizing generative audio, making high-quality AI audio tools more accessible.

With its focus on fast, local processing and open licensing, Stable Audio Open Small stands out in a market dominated by cloud-based solutions.

News Gist

Stability AI, in partnership with Arm, released Stable Audio Open Small—a lightweight, open-source AI model that turns text into short audio clips.

Optimized for speed and local use on Arm CPUs, it supports real-time generation.

Free for most users, it aims to make generative audio tools widely accessible.