New AI can generate full three-minute songs – Kommersant

New AI can generate full three-minute songs - Kommersant


Stability AI presented a new neural network Stable Audio 2.0, which will allow users to upload tracks into it, on the basis of which the AI ​​will create a full-fledged song with a classic structure: intro, development and ending. You can try the neural network for free on the project’s official website or via the API.

Stable Audio 2.0 was trained on more than 800 thousand tracks. According to the company, artists could refuse to provide their compositions for training the neural network. It will be possible to adjust the degree of similarity to downloaded tracks. Those who tried the neural network received mixed impressions: the vocals do not resemble human ones, and sometimes the output is very strange compositions. The original version of the neural network was launched in September 2023 and was able to create audio files no longer than 90 seconds.

Formerly the advocacy group Artist Rights Alliance published An open letter from more than 200 music artists, in which industry representatives expressed concern about the misuse of their music to train neural networks, and also called for speedy regulation of this area. Signatories included Billie Eilish, Stevie Wonder, Katy Perry, REM, Pearl Jam and others.

Evgeniy Fedunenko


Source link