It shouldn’t be a surprise to see Warner Music boss Robert Kyncl praising YouTube: he did work there for 12 years before moving to the major label after all.
However, Kyncl’s warm words for his former employer, made during WMG’s latest quarterly earnings call, are still important. They focused on YouTube’s freshly-announced AI music experiments in partnership with artists.
“I’d like to actually point out to the significance of this,” said Kyncl. “Imagine in early 2000s, if the file sharing companies came to the music industry and said ‘would you like to experiment with this new tool that we built and see how it impacts the industry and how we can work together?’ It would have been incredible.”
“Obviously, that didn’t happen. So this is the first time that a large platform at a massive scale, that has new tools at his disposal is proactively reaching out to its partners to test and learn.”
Kyncl also broke down WMG’s three core priorities around AI technologies. First, to ensure that AI firms are licensing content properly for training, keeping records of their inputs and watermarking any content that their models generate.
Second, it wants the platforms where this content will be distributed – from YouTube, TikTok and Instagram to Spotify – to support the music industry’s desire for “control, attribution and monetisation… we have a blueprint from all of our work on user-generated content over the past 15 years or so”.
Third, WMG is pressing policymakers – both through its work with industry bodies and on its own – to craft laws ensuring that “licensing for training is required and also that name image likeness and voice is afforded the same protection as copyright”.
Talking of watermarking, that cropped up in a blog post yesterday by DeepMind, Google’s AI division, whose technology is being used for YouTube’s first experiments with AI music tools and artist voice-clones.
The blog post provided more details on the Lyria model that is being used. DeepMind said that it “excels at generating high-quality music with instrumentals and vocals, performing transformation and continuation tasks, and giving users more nuanced control of the output’s style and performance”.
All content created using Lyria will be watermarked with a tool called SynthID, which is inaudible to the human ear, but “designed to maintain detectability even when the audio content undergoes many common modifications such as noise additions, MP3 compression, or speeding up and slowing down the track”.
This kind of detectability is something that music rightsholders have been calling for – including Kyncl yesterday.
One thing missing from DeepMind’s blog post is information on what music Lyria was trained on, although we’d imagine that information will have been shared with YouTube’s rightsholder partners, given their keen interest in the matter.