Brief Summary

The tech behind AI is developing so fast that the music industry and the legal system is getting left behind. AI generated vocals of well known artists are just one of the challenges we’re all going to have to catch up with.

Going Deeper

It’s less than a year since the, widely ridiculed, story of an AI generated rapper being dropped by its record company for creating controversy over racial stereotyping and its use of unacceptable language was widely reported in the media

As amusing as it was, the development and refinement of the technology behind FN Meka, the virtual artist created in 2019 which gained such notoriety, was inevitable and the speed at which AI generated vocals has progressed has been remarkable. The best of these are now capable of recreating voices of artists which are extremely familiar to the listening public and this opens up an enormous can of virtual worms for the music business which will inevitably intensify over the coming years.

On our recent podcast Things We loved In April 2023, embedded below, as her find of the week Ashea featured a news story about the recording artist Grimes, who had announced that she was offering use of her AI voice to anyone who wished to use it in exchange for a 50% cut. The first thought which occurred to me was whether she was offering use of something she didn’t actually own? We’ll return to that. My second thought was that I’d prefer to use John Lennon’s voice on my next recording. Inevitably my glib comment indicated my ignorance of the situation in this dizzyingly fast moving area as people have already started. I’m sure there’s a Lennon out there already but check out this AI Liam Gallagher. This is an album of original material recorded by a real band, amusingly names “AIsis” with an AI Liam on vocals.

Wow! So what does this mean to the music business? Use of AI to generate vocals for music certainly raises a host of legal, ethical, and artistic issues.

Will AI Vocals Benefit The Recording Industry?

Staying on the positive, the ability to synthesise new audio which is recognisable as, or indistinguishable from, the singing voice of a known artist will have innocent uses in restoration, harmony and backing vocal generation and the like. Inevitably collaborations with retired or dead performers would also be overwhelmingly tempting. For record companies who are restricted by a finite amount of recorded legacy material, a posthumous release of original material is irresistible, though there is always the danger of such behaviour being seen as ghoulish. However the listening public are likely to become desensitised to such concerns.

Moving beyond the ethical concerns around using deceased artists, ageing artists whose voices have changed since their youth might welcome such treatment. Paul McCartney can’t match his voice as it was in his 20s, though in a video recently removed from YouTube, AI was used to take the original of ‘I Don’t Know’, released when Paul was in his mid-seventies to create the sound of a much younger McCartney.

The extraordinary ABBA live event Voyage uses visual avatars of ABBA in their prime. If the visual likeness, why not the voices too?

The use of that word ‘likeness’ is significant here as in an area as fast moving as AI the slow moving legal world is unlikely to catch up anytime soon. Copyright and publishing protect recordings and the content but the voice itself is neither of these things. I’m not a lawyer so I’m not going to offer any legal opinion here but I can certainly see some protracted legal wrangling on the horizon.

If they aren’t already, I’d expect record company contracts to be including clauses covering this very soon but we are definitely in a ‘Wild West’ phase of a new technology here and it’s going to be interesting for the next few years.

What Are The Challenges Presented By AI Vocals?

• Legal - The issue of copyright ownership. If somebody uses an AI model to generate vocals indistinguishable from a well known artist, who does it belong to?

• Ethical - Beyond the concerns of the Artist, or the estate of the artist in the case of a deceased artist, what about the audience? Should an artificial vocal be clearly identified as such? Does it matter and do the audience care? Is this different from a string section being created by a Virtual instrument rather than real players or is the vocal different in some way? If so, why?

• Artistic - Will AI-generated vocals lead to a homogenization of sounds if producers and artists rely on algorithms to generate vocalizations? In the same way as some criticise tuning algorithms for creating a ‘sameness’ pervading some styles of music, might AI create a similar convergence? What about about the role of creativity in the production process? Might the use of AI result in an even more formulaic approach to making music?

• Authenticity - Finally, there are questions about the quality of AI-generated vocals, and whether they can ever truly replicate the nuances and emotions of human singing. This can get deep but if an algorithm conveys an emotion it is incapable of experiencing, can it have value? At the rate AI is progressing the algorithms might well be experiencing emotions for real before long…

Every new technology brings both a positive and a negative reaction. Are AI vocals really any different from synthesisers, sampling, Beat Detective, Auto Tune? It raises a question about what really is unique? In another recent podcast Acon Digital’s Stian brought a research paper on Music LM, an example of AI generated music. If the music can be written by AI, and the instruments and vocals can be generated by AI then it’s going to be an interesting few years. I’m sure I won’t be listening to the inevitable waves of derivative 100% AI music being churned out. I’ll be listening to all those suspicious new Beatles albums from the 70’s and 80’s!

Photo by Pixabay