Stable AI startup, known for its II Generator of the images of Stable Diffusion, presented a new open model for creating sounds and songs. This model, named Stable Audio Open, studied exclusively on records without copyright.
Stable Audio Open allows you to create musical notes lasting up to 47 seconds based on a text description, for example: “Rock Bit played in a processed studio, a session eardrum on an acoustic installation.” The model was trained in ~ 486,000 samples from libraries Freesound and Free Music Archive.
The model can be used to create drum bits, instrumental riffs, background noise and “production elements” for videos, films and television shows, as well as to “edit” existing songs or apply the style of one song to another.
“The key advantage of this open release is the opportunity to configure the model for their own audio-domed ones,” writes Stability AI in Corporate blog.
However, Stable Audio Open has restrictions. She cannot create full -fledged songs, melodies or vocals – at least high -quality. Stability AI claims that the model is not optimized for this, and offers users who seek such opportunities to use the paid service Stable Audio.
In addition, Stable Audio Open cannot be used for commercial purposes; Service conditions prohibit this. The model also does not work equally well with all musical styles and cultures, as well as with descriptions in other languages except English. Stability Ai explains this with the bias of training data.
Stability AI, which has long been trying to improve its financial situation, recently found herself at the center of the scandal after the Vice President of Generative Audio, Ed Newton-Rex, resigned due to disagreements with the position of the company that Education of generative AI models on copyright -protected works is “conscientious use.” It seems that the release of Stable Audio Open is an attempt to change this narrative, while advertising paid products Stable Ai.
With the growth of the popularity of musical generators, including Stable Ai products, copyright and possible abuse by some developers become the main topic for discussion.
In May of this year, Sony Music, representing artists such as Billy Joel, Doja Cat and Lil Nas X, sent a letter to 700 companies engaged in AI with a warning against the “unresolved use” of their content for training audio generators. Meanwhile, in March, in the American state of Tennessee, the first law in the United States