The neural network trained on different genres.
The developers of OpenAI labs introduced the artificial intelligence of Jukebox, which itself composes music, text and writes songs. As reported by The Next Web, AI is trained in different genres: it “listens” for a particular musician, analyzes words, and then simulates his style, writes knife.media.
Read more about Jukebox is written in the OpenAI blog, there you can find the link to all the tracks recorded algorithms. As can be seen in the impressive discography, the AI tried himself in the folk, jazz, hip-hop, indie rock and other genres, including 22 species of metal. For example, he sang the voice of Kanye West, Michael Jackson, Elvis Presley, and singer of Megadeth.
Just the model of “studied” 1.2 million songs, each of which algorithms have identified design features of music and lyrics. Happened, however, is not without nuance: the authors of TechCrunch comparing the tracks with a drunken karaoke, audible through the fog of narcotic intoxication, because musical genres are sometimes changed on the go, text is often not apart, and if apart, it can break the brain due to the lack of logical connection between words.
In General, the algorithms do their job well: in the playlist of the most successful country music songs sounds really like a country.
Unfortunately, you cannot test online Jukebox, as it was with the bot Talk to the Transformer from the OpenAI, but at your own risk, you can explore extensive work with AI. According to the developers, the songs will be enough for 10 days of continuous listening.