Fake Features: Is There a Way to Legally Use AI Generated Vocals in Music? (+ FAQ)Fake Features: Is There a Way to Legally Use AI Generated Vocals in Music? (+ FAQ)


Fake Features: Is There a Way to Legally Use AI Generated Vocals in Music? (+ FAQ)

Jay-Z got 99 problems and now AI is one. AllttA's "Savages" shook the hip-hop world with its AI generated Jay-Z feature. At first listen, the custom voice is hard to distinguish from the real Hova. The release—produced with a Tracklib sample—opened up a conversation about the ethics of using generative AI in music. Producers including Pete Rock, 9th Wonder, Lord Finesse, and Young Guru all weighed in. But what do these rapid-paced, groundbreaking developments hold for the future? And can you actually legally use or sample AI generated vocals?


Danny Veekens


January 30, 2024

Click here to jump straight to the FAQ, answered by sample clearance specialist Deborah Mannis-Gardner.

"Savages" was released as "a try, an experiment, an experience," explains AllttA emcee Mr. J Medeiros: "It's not monetized, and it's not officially out. (...) We felt it was 'important' enough of an experience to share outside the secrets of our own studio. The motivation to put it out was to enter this discussion [on AI] happening now from a musical perspective."

For the beat, AllttA producer 20syl looped "A Wish To A Star" by composer Willis Schaefer, taken from the Cavendish Archive library catalog on Tracklib. "I was trying to recreate my late-90s production style for this track which was originally intended for a documentary soundtrack," says 20syl. "Back in the 90s, I used a lot of movies soundtrack from Italian and French composers with orchestral arrangements. When I heard 'A Wish To a Star,' the section I looped grabbed my attention because it had the nostalgic atmosphere and texture that I was looking for. I used my old MPC 2000 XL to sequence the drums, combined with the sample in Logic Pro, and I recorded the bassline on my Fender Precision."

Custom voices

The AI-generated Jay-Z was added to "Savages" two years ago, with a text-to-speech online tool called Uberduck. Their voice generator software offers an option modeled after famous rappers with a custom voice creator. "There were a lot of rapper options but Jay-Z seemed to work best, possibly given that Jay-Z and I share a similar tone or cadence at times," says Mr. J Medeiros. "Frankly, the models in that period were not very good—as it was so new but not clean enough to pass by the trained ear. It felt like sampling for the first time again. (...) Or taking drum machines and sample time lengths to new levels by creatively tweaking its official 'means.'"

Voice cloning

But unlike the real Jay-Z—who is rumored to record verses in one take,—working with the 'fake Jay-Z' wasn't easy… Medeiros: "It took a month at least to get the verse to sound right, between 20Syl engineering the vocals over and over again and tricking the AI to say the words correctly, through misspellings and different sounds and vowels. We changed a lot of the text to get the AI to say the words correctly. We exported a lot of the verses, one or two lines at a time, until the flow and wording were good enough."

"It felt like sampling for the first time again. Or taking drum machines and sample time lengths to new levels by creatively tweaking its official 'means.'"

—Mr. J Medeiros (of AllttA)

The heated debate about AI generated vocals

The online debate about the voice cloning on "Savages" really took off once YouTuber Marques Brownlee uploaded one of his explainer videos, and when engineer and long-time Jay-Z collaborator Young Guru shared "Savages" on his Instagram, saying, "I've been trying to tell everyone that this is where we are now with AI. (...) On one hand, I'm well aware that you can't stop technology. On the other hand, we have to protect the rights of the artist. Not only artist [sic] but everyone in society. People should not be able to take your Name, Image and Likeness without permission. We have to add the voice to this law."

Young Guru refers to the legislation of the right of publicity to control the commercial value of one's name, image, likeness, signature, and other personal traits—which already includes the voice. Unlike creative works that are protected by copyright law, someone's identity isn't a "creative work." But that doesn't mean it's not protected: similarly to copyright, personal identity is also protected from freely being used commercially without consent. Especially in the case of celebrities, although actual laws and limits to the right of publicity differ per country and even per U.S. state.

One of the explainer videos:

Past lawsuits about voice clones

Even with the fast-paced developments in generative AI, the ethics and legal issues around mimicking someone's voice are not new. This even goes back decades: in 1988, actress and singer Bete Midler filed—and won—a lawsuit against Ford Motor Co. and advertising agency Young & Rubicam, who used ten voice clones for an ad campaign. "A voice is as distinctive and personal as a face," the appeals court ruled. "When a distinctive voice of a professional singer is widely known and is deliberately imitated in order to sell a product, the sellers have appropriated what is not theirs."

A more recent case involves Rick And Morty voice actor Jess Harnell, whose metal mash-up cover of Journey's "Don't Stop Believin'" was so spot-on that Steve Perry, the lead singer of the band Journey, thought they had "sampled his voice without permission." Perry won the case after the judge ruled the song was "misleading to the public."

Universal Music Group's sledgehammer response

After a deep-fake collab between Drake and The Weeknd called "Heart On My Sleeve" surfaced on DSPs last week, the shit really hit the fan. The song, made by a mysterious entity named Ghostwriter, went viral within a few days time before Universal Music Group (UMG) went all-out war mode and had the music taken down on all platforms.

"Platforms have a fundamental legal and ethical responsibility to prevent the use of their services in ways that harm artists," UMG told Billboard in a formal statement. "The training of generative AI using our artists' music (which represents both a breach of our agreements and a violation of copyright law) as well as the availability of infringing content created with generative AI on DSPs, begs the question as to which side of history all stakeholders in the music ecosystem want to be on."

What's next for AI generated vocals?

Viral examples like "Savages" and "Heart On My Sleeve" set a precedent for the future of using speech tools and natural-sounding voices. When a voice is recognizable, that touches on two different types of protection: the right of publicity and—if used to train generative AI—copyright of the original voice. "It seems inevitable the issue will wind up in courts," Variety writes in response to Universal Music Group's public statement.

But there's also another side to all of this. Late last year, James Earl Jones (the original Star Wars voice actor of Darth Vader) licensed his voice to an AI company. Earlier this week, musician Grimes invited people to use AI versions of her voice, offering a 50% royalty split. With potential revenue streams on the horizon, it's a matter of time before the first rappers or hip-hop moguls start cashing in on the current trend of using AI generated vocals in music. Tracklib's bet is that it's bound to happen soon.

Similar to the early days of sampling with free-for-all unrestrained creativity with a lack of the right legislation and precedents for rightsholders to keep control over the use of their copyrighted works, the same now goes for the use of artificial intelligence in music. But just like sampling lawsuits such as Biz Markie VS. Gilbert O'Sullivan, it's a matter of time before the first industry-changing case on generative AI in music. "We don't have the answers to the many questions the song ['Savages'] or AI generally bring up," Mr. J Medeiros adds. "We may have one answer, at least: do we have a right to 'try' this new tech that has landed in our lap? Yes, we think so."

FAQ: Deborah Mannis-Gardner on AI-Generated Vocals

deborah mannis gardner samples ai vocals

Do you see any parallels between the early days of sampling and the current experimental phase of using AI generated vocals in music?

I see huge similarities. Let’s not forget it’s a human programming the AI content. And to train it to sing in a certain style, they need to feed it music featuring vocals in that style. There is a need to establish a value for the original rightsholders so this does not turn into Napster with labels and publishers calling to shut everything down. Like sampling, there should be a share in the revenue made from these AI sources.

How do you think old copyright laws are going to catch up with this new voice synthesis technology?

Old copyright laws haven’t even caught up with sampling. They are still struggling to accommodate DPS, so I don’t foresee this in the near future, but eventually, I would like to believe they will.

What if a vocal flow is derivative of, say, Jay-Z, but tweaked enough for it not to be distinctively recognizable and thus not protected by the right of publicity (and also not copyrightable in the case of generative algorithms)?

This is a debate we are going to need to have very soon, and I have already heard a variety of arguments. Similar to Web3, the current legal framework makes it difficult to address AI because it didn’t anticipate this, so I believe we need to come up with new solutions.

Perhaps a discussion needs to be had on how an artist, producer, or writer could define their style and technique as a "Profile" from which AI may be collecting, and then use that to provide protection for the artist/creative’s uniqueness? It’s still early days, but we need to start thinking outside the box.

The million-dollar question for producers at the moment: is there a way to legally use AI generated vocals?

If a producer trains an AI to sound like a specific artist/producer/writer, to me that is willful infringement hiding behind the argument of AI. This would require the use of music from that artist to train the AI properly, and licenses should be required for that use, just like sampling. This is merely my opinion.

Keep digging

Show more


We are unfortunately unable to offer support in the comments. If you have any questions, reach out to us here!