Join our newsletters for daily and weekly updates, providing the latest insights and exclusive content on leading AI advancements.
Stability AI is expanding its generative AI capabilities for audio with the launch of Stable Audio Open 1.0. Known for its text-to-image generation technology, Stability AI also offers various models for code, text, and audio. In September 2023, they introduced Stable Audio, a text-to-audio generative AI tool. On April 3, Stable Audio 2.0 was released, offering improved clarity and longer audio generation.
While the Stable Audio tool can generate up to 3-minute audio tracks for commercial use, the new Stable Audio Open is more limited. It’s designed for shorter audio pieces like sound effects rather than full songs. Stable Audio Open is accessible under the Stability AI non-commercial research community agreement license, which grants open access but limits its use.
“Our aim with Stable Audio Open is to give audio researchers and producers access to one of our generative audio models to boost research, adoption, and creative use of these tools,” said Zach Evans, head of audio research at Stability AI.
Stable Audio Open is optimized for creating drum beats, instrument riffs, ambient sounds, and other audio samples for music and sound design. Unlike the commercial version that produces tracks up to three minutes long, Stable Audio Open generates high-quality audio clips up to 47 seconds using text prompts.
Stability AI trained the model using audio data from FreeSound and the Free Music Archive, avoiding any unapproved use of copyrighted material.
One key advantage of Stable Audio Open is that users can fine-tune the model with their own audio data. For example, a drummer could customize the model with their own drum samples to create unique beats. Fine-tuning is supported through the Stable Audio Tools library, which is openly licensed, and the Stable Audio Open Model weights are available on Hugging Face.
“The audio research team is continuously working to enhance the quality and control of our generative audio models,” Evans stated. “We anticipate more releases, both commercial and open, showcasing our research progress.”