Legality of AI vocals from royalty-free music

Question

I'm a fairly new music producer, and one of the inherent problems with it is that it means you have to do all the work yourself: sound design, any vocals, mixing, and mastering. There's quite a few companies that make this easier by providing royalty-free samples. Sometimes it's a free pack being given away, sometimes it's a one-time purchase. Problem is, samples are set in stone, what you download is what you get. This is a problem with vocals, because you can't really add any more words to them, or anything like that.

With that said, there are a growing number of websites that use AI and text-to-speech to create audio files in a pre-chosen voice. Many of these also feature a way to create a new "voice" by providing some sample audio. So one way around this would be to take a bunch of samples with a voice you like, and use them to create a new AI voice; you could then use it to make any samples you want.

My question is, is this legal? My assumption is that if a sample is being offered royalty-free, then the vocalist had to have agreed to a one-time payment. And it's pretty routine to cut up these samples in an arrangement, in addition to manipulating the sound of the voice itself. If the creation of the AI and production of new samples is this same kind of manipulation, it seems like it should be fine. On the other hand, I could see an argument that this is just stealing the voice of the artist. Extra question, if you were mangling the voice so much it didn't even sound like the original anymore, would that change things?

Answer 1

The contemplated use is probably not "fair use" which can be done notwithstanding the copyright in the samples. But the license appears to allow for derivative works, so it should be permissible.

A right of publicity in the voice itself, normally doesn't apply when the model's voice is used with the model's permission for compensation.

Answer 2

A license really should address this question now days, but many licenses still don't.

So you're doing a thing where there is possibly going to be a dispute as to whether or not the license covers the thing. You are going to have an argument for why the thing is so much like the things that the licensor actually intended to cover that it should count. Maybe the license was clever enough to include terms about media or forms of processing not yet invented, and whether they are or are not included as a group. But the licensor might have an argument for why this new thing you want to do that they didn't know was possible when they drafted the license is not actually a thing you had their permission to do. And you might end up fighting it out in court.

If you can contact the licensor and come to an agreement with them about whether this is or isn't meant to be covered, it might save a lot of trouble later.

Alternatively, if the final song you put out doesn't sound anything like the samples that were processed by your algorithm to produce the new sounds you actually used (and you don't just have what is clearly Ash the vocalist who did the samples singing a song they never agreed to sing), the person with copyright on the sample library might not feel inspired to bring a copyright case. How would they ever know you ever did the processing, in order to form the opinion that it would not have been covered by the license?