Stability AI has introduced a new open-source AI model named Stable Audio Open, designed to generate sounds and songs.
The model was trained using approximately 486,000 royalty-free recordings sourced from FreeSound and the Free Music Archive.
Stable Audio Open can create audio recordings up to 47 seconds long based on text descriptions such as “Rock beat played in a treated studio.”
This AI model is capable of producing various sounds, including drum beats, instrument riffs, and ambient noises, making it useful for multimedia projects. It can also edit existing songs or apply the style of one song to another.
Users can fine-tune the model with their own audio data, allowing drummers, for example, to create new beats using their recordings.
However, Stable Audio Open has some limitations.
It cannot effectively produce full songs, melodies, or vocals as it is not optimized for these tasks.
Additionally, it cannot be used for commercial purposes as per its terms of service.
The model’s performance may also vary across different musical styles and languages due to biases in the training data.
The release of Stable Audio Open comes amid some controversy.
Stability AI’s VP of generative audio recently resigned over disagreements about training AI on copyrighted works.
The introduction of this open-source model may be part of an effort to improve the company’s image while promoting its paid services.
