Google and Universal Music reach agreement over AI “deepfakes”

Google and Universal Music have begun discussions to license the melodies and vocals of artists for songs created by artificial intelligence. The music industry is trying to monetise its biggest threat.

Four people with knowledge of the situation confirmed that the discussions aim to create a partnership with an industry grappling with new AI technologies.

As a result of the rise of generative artificial intelligence, “deepfake songs have become increasingly popular. They can be convincingly mimicked, and often without consent, by established artists.

Frank Sinatra’s vocals have been used in a hip-hop version of “Gangsta’s Paradise”, while Johnny Cash’s was used for the pop song “Barbie Girl”. PluggingAI is a YouTube user who offers songs that mimic the voices of deceased rappers Tupac, Notorious B.I.G.

Jeffrey Harleston, general counsel at Universal Music, told US lawmakers that stealing an artist’s voice, regardless of the method, was wrong.

The discussions between Google and Universal Music have not yet reached a conclusion and there is no imminent product launch. However, the goal is to provide a tool that allows fans to legally create tracks and to pay the owners of copyrights. The people close to the situation said that artists would be able to opt in.

A person with knowledge of the situation said that Warner Music, third largest music label in the world, was also talking to Google about an upcoming product.

Music executives compare the rise of AI generated songs to the early days of Google’s YouTube, when users started adding popular songs to their videos. After years of battling YouTube over copyright violations, the two parties have now established a system whereby the music industry receives about 2bn per year in return for user-generated videos.

AI is gaining popularity, and some stars are worried that their songs or voices will be replaced by fake versions.

The issue came to light earlier this year, when a song produced by AI that mimicked Drake and The Weeknd’s voices went viral on the internet. Universal Music, the label that represents Drake, Taylor Swift, and many other well-known musicians, removed the song from streaming platforms due to copyright violations.

Drake in April called it the “final straw” when another song mimicked his voice using AI. Rapper Ice Cube described such tracks as “demonic”.

Other artists have also embraced this technology. Grimes has agreed to split royalties with people who use her voice for AI-generated music.

She told Wired Magazine this week that AI tracks that use her voice are “good stuff”. “They were so similar to what I might do with my next album that they were disturbing,” she told Wired magazine this week. . . It’s a bit like saying, “Oh, sick, maybe I get to live forever.” I’m into the self-replication.”

Robert Kyncl is the chief executive officer of Warner Music. He told investors on Tuesday that AI, “with the proper framework in place”, could “enable users to give their heroes an ultimate compliment by creating a whole new level of user generated content. . . “Including new cover versions and remixes”.

He said that artists should be able to choose whether or not they want to participate. He said that some people may not like the idea, which is fine.

Google’s music product would help it compete with competitors such as Microsoft. Microsoft has invested $10bn into OpenAI, the leading AI company, and owner of GPT-4, the industry-leading AI model.

Google’s AI-based products, such as Bard the chatbot, have been launched to compete with Microsoft’s Bing software and search engine.

In April, Universal Music urged streaming platforms not to allow AI services to scrape their songs without permission. The company, which controls a third or more of the global market for music, has asked Spotify and Apple not to allow developers to use its catalogue to train AI technologies.

According to sources familiar with the matter, Lyor Cohen is a former executive of a record label who now leads YouTube’s Music Division. He has worked on this project for Google.

Google released a preview of AI-powered software for music in January. The software was able generate music based on text descriptions like “upbeat arcade games” or “reggaeton fused electronic dance”, and even more specific prompts such as “a calming melody with a distorted guitar chord”.

The authors of the article pointed out that the software could potentially violate copyright laws when it reproduced music by specific artists from its training data. Google, however, released its experimental MusicLM tool to consumers in May. It said that it worked with artists on the development of it.

Google, Universal Music, and Warner Music declined comment.