Stability AI, a leading company in the field of artificial intelligence, has announced the release of a new model for generating open-source sounds, named “Stable Audio Open”.
The company stated in its official statement that this model was trained using recordings that are free of copyrights, based on approximately 486 thousand samples from open-source sound libraries such as FreeSound and Free Music Archive.
The “Stable Audio Open” model works on converting texts into audio recordings up to 47 seconds long. The company claims that this model can be used to create sounds for various purposes, including films and TV shows.
Stability AI pointed out that one of the main benefits of this open-source version is the ability for users to enhance the model by using their own personal audio data. For instance, users can improve the model by using samples from their own recordings to create new sounds based on them.
The company advises users in need of increased sound production capabilities to take advantage of its “Stable Audio” service.
The “Stable Audio Open” model is not suitable for commercial use according to its terms of service, and it does not function efficiently with script claims in languages other than English or when generating audio clips that are not compatible with the training data.
Stability AI has faced challenges in recent times in developing its products, especially after the resignation of the Vice President of Generative Audio, Ed Newton-Rex, who claimed that the company was training generative AI models on copyrighted works.
Through the launch of this new model, the company aims to protect itself from the accusations leveled against it and enhance its advanced services in AI-based sound production.
In May, Sony sent a letter to hundreds of companies in the generative AI field warning them of the “unlawful use” of their content to train sound generators.
