Google is testing a feature that will allow users to create musical Shorts with AI-generated voices of famous artists. The singers whose voices are imitated by Google’s Deepmind tool must first give permission for this.
The feature, Dream Track, initially makes it possible to use the voices of nine artists for YouTube Shorts. This includes Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Papoose, Sia, T-Pain and Troye Sivan. This feature can generate a song of up to thirty seconds, including lyrics, a backing track, and the AI voice, based on a short text prompt.
Based on the type of song users describe in the prompt, it automatically matches an artist that best matches their desired music style. For example, in Google’s demonstration, the command is “A sunny morning in Floridia, R&B.” The song generated from it will automatically include T-Pain’s AI voice. This feature is initially only available to “a small group of American creators,” Google says.
In addition, Google is coming with some other music-related AI tools for YouTube. These should make it possible to create backing tracks in an instant. In this way, the company shows that it is possible to make music by just humming the melody and describing in a text prompt which musical instruments should be used. Monophonic singing can also be turned into an orchestra, or a midi keyboard melody into singing. According to Google, these tools can be tested later this year by participants in its Music AI Incubator .
The tools all come from Google’s Deepmind division. The machine learning model Lyria was used for this, which is specifically trained for generating music. The company states that tracks generated with this AI model will automatically contain a watermark, so it should be possible to find out at any time whether Lyria has been used.