Though Stability AI has recently garnered much of its fame from its extremely popular image generation model, Stable Diffusion, the firm released an extremely cool new audio-generating model called Stable Audio Open Small. This announcement comes as the company continues to chart its path through a stormy chapter filled with executive turnover and financial reorganization. The audio model is meant to be computationally efficient enough to run directly on smartphones, making advanced sound capabilities not just powerful, but broadly accessible to all users.
In fairness, in recent months, Stability AI has taken some lumps in the press. Co-founder and now ex-CEO, Emad Mostaque, has been accused of running the company into the ground and severely mismanaging, crashing the company’s finances. In a bid to turn the situation around, Stability AI raised new funding last year, attracting investors eager to revitalize the beleaguered firm. To solidify their own stability, Stability AI recently added new CEO Emad Mostaque, naturalizing some fresh leadership to the generative AI hype factory.
Adding to the company’s notable changes, renowned filmmaker James Cameron has joined Stability AI’s board of directors. His deep expertise and visionary outlook on the future will no doubt bring invaluable insights as the firm pushes boundaries in this nascent space of artificial intelligence. Stability AI has released several new image generation models over the past few months, demonstrating its ongoing commitment to advancing AI technology.
Among those that were recently released, Stable Audio Open Small has the most parameters at an impressive 341 million. It’s tuned to perform exceptionally well on Arm CPUs. Using this model on-device enables real-time generation of up to 11-second-long audio clips in under eight seconds on a smartphone. Stable Audio Open Small is good for efficiently generating brief audio clips and sound effects. It focuses on efficiency and speed to help streamline your creative workflow.
Please note, the audio model only supports prompts in English at this time. Furthermore, it no longer affords the production of naturalistic singing or the whirlwind artistry of a great song. Despite these limitations, Stability AI offers Stable Audio Open Small free of charge for researchers, hobbyists, and businesses with annual revenues below $1 million. For any developers and organizations with revenue above that limit, an enterprise license will be needed.
In its defense, Stability AI points to how it exclusively adopts training sets devoid of copyrighted material. This model significantly reduces the IP risk vs alternative models on the market. This user-centric approach contributes to a safer environment for users. It introduces them to the possibilities of audio generation without the threat of copyright infringement hanging over their heads.
