Do you own your voice? The challenges of voice cloning

A couple of weeks ago I was discussing in class how some comedians cloned George Carlin‘s voice, and the subsequent lawsuit by his estate. I won’t go into any details, Aaron Moss has written the best analysis of the merits of the case, but suffice it to say that it involves the use of voice recordings to train an AI that sounds like the late comedian, and it was used to record a comedy special. The discussion in class was centred on the legal remedies available to fight voice cloning. At some point, I commented to the class: “there are hundreds of hours of me speaking, I wonder if anyone has ever cloned my voice”. The students took up the challenge, and made a clone of my voice using class recordings.

Hilarity ensued.

Evidently I’m fine with this as it is a direct response to a discussion in class, and it was a very nice use of my voice, as well as being a great didactic resource. It was also rather disconcerting, the recording sounds a bit like me, even with some of the hesitancy that is characteristic of my speech patterns, but it clearly spoke faster, and the accent wasn’t entirely right. All in all, this is one of the reasons why I love teaching, I’m endlessly surprised by my students.

But evidently this got me thinking about possible negatives, what if my voice was cloned for nefarious purposes? This also should be a warning to other academics, our voices are recorded every day in lecture theatres and seminar rooms, so the technology has gotten to the point where we will have to start thinking about the implications of this.

So can existing law be used to stop an unwanted voice clone? The answer is complicated. I won’t deal whatsoever with privacy and data protection, which could potentially be used for this, I will try to concentrate on UK IP law in general.

Do you own your voice?

The short answer is no, there’s no voice likeness right in the UK, unlike other countries which protect this, such as publicity rights in the US.

So, are there any other ways you can use to stop voice cloning?

Copyright

As voice cloning has to be done with a recording, the most logical place to start would be by trying to enforce the copyright on the actual recording used for training the AI to generate a model of your voice.

However, this only works if you own the sound recording of your voice, which may not be the case, this is a similar problem with photography, where the person depicted in the photograph doesn’t own the copyright, the photographer does. In the case of a sound recording under UK copyright law, the owner is the producer, which is usually defined as the person who made it possible for the recording to be made. This can be straightforward most of the times, the person making the recording is the producer and therefore the owner, with no rights going to the person being recorded. That is, unless the  person being recorded has rights arising from their own intellectual property which could stop someone from making a recording, I’ll deal with that a bit later.

In the case of the students who made the training of my voice, there are two possible sources for such a recording. The first would be their own recording in class, I allow this as most of my students at postgraduate level are non-native English speakers. In that case, they own the recording, and therefore can do with it as they please, potentially even using it for training an AI without my permission. I would need other rights in order to stop that from happening. Another common source of recording in academia is through lecture recording, to make a long story short, nowadays almost all of our lectures are recorded, and these are made available to the students enrolled in the module. I don’t own those recordings, they’re technically owned by the university, so these would be work for hire. But as students have the right to access and use those copies, does that mean they’re precluded from making use of the recordings to train an AI? I don’t think so, because training is not one of the exclusive rights of the author, unless you consider making an AI voice clone an adaptation of the work under copyright (a derivative in other jurisdictions), which I don’t think it is.

There is potential for protecting actual conversations and speeches under copyright, but this depends on two factors, it has to be fixed, and it has to be original. A lecture has a higher likelihood of being protected as it is often a more structured piece of speech, often following prepared notes, slides, and also reading out from those materials, so there could definitely be protected. The owner of my university lectures, however, is the university. As per conversations and seminars, this is more difficult, MacQueen comments that “With conversation or dialogue between two or more people, further issues may arise from a lack of ‘distinct, identifiable boundaries’ between the various contributions”. However, “judges have been prepared to accept that, viewed as a whole, such dialogues can give rise to copyright works, and also that particular quotations from the larger whole of the dialogue can have meaning even taken in isolation, which is what made the act of quotation worthwhile in the first place.”

But again, this is the protection of the underlying conversation, not the actual voice, which is what is being trained. It is important to make a distinction between the rights on the sound recording itself, and any other underlying rights on what is being recorded. So for example, the recording may be of a singer performing a song (more on that later), and there are rights on the music and the lyrics.

So let’s imagine that someone is trying to replicate a singer’s voice, while the voice itself has no copyright, the owner of the sound recording may be able to sue the person making the training for copyright infringement, but that would only work if in order to make the voice clone the person making it has made copies of the sound recording without permission. The situation would be trickier if the person was using a legally purchased copy of the song and making a transient copy out of that copy in order to train the AI. The problem is that, as I have mentioned before, training an AI is not an exclusive right of the author.

To summarise, relying on copyright of the sound recording, and even the underlying speech as a literary or dramatic work, would be unreliable.

Performers’ rights

Another possible avenue for protection is Performers’ Rights, this does exactly what it says on the tin, these are related rights that protect performances. According to the CDPA, a performance is “(a)a dramatic performance (which includes dance and mime), (b)a musical performance, (c)a reading or recitation of a literary work, or (d)a performance of a variety act or any similar presentation”. This covers evidently singing by musicians, as well as acting by actors, but it also covers a wide range of other public acts that could fall under the definition of reading a literary work, as well as speaking in a dramatic manner. I do talk a lot with my hands, is that a dramatic work? I digress.

So the performer´s rights are widely applied, and could cover voice clones. Talking about deepfakes in general, Pavis commented:

“Importantly, the law makes no reference to notions of originality, quality, aesthetic merit, professional or amateur status in defining protected performances or performers. Similarly, there is no requirement based on the performer’s reputation, notoriety, professional or amateur status, or goodwill to receive protection under performers’ rights. Under this definition, teachers, politicians, journalists as well as individuals sharing their videos on social media are ‘performers’ legally entitled to control the records made of their ‘performances’. As a result, the use of these recordings as input data to generate a Deepfake would fall within the scope of performers’ rights granting protection to the target and driving individual (if adequately reformed)”.

The rights granted to a performer are similar to the exclusive rights under copyright, with a few added ones. So a performer has the exclusive right to stop people from making a recording of the performance, as well as a right to equitable remuneration from the commercial exploitation of a performance; and also the right to stop the copying, lending, publishing, and communicating to the public the performance. It must be mentioned that there is no right of adaptation.

So assuming one does not have permission to make a recording of a performance, it would be possible to object to the making of a copy to make a voice clone. Conversely, if one has permission to record the performance (as I have done with my classes), I would find it difficult to stop someone from using the recording to make a voice clone.

Passing off

The Tort of passing off could be used to protect against a voice clone, but this one is more difficult. In order to be able to claim passing off, there are three main requirements, the subject must have goodwill, there has to be a misrepresentation, and there have to be damages arising from that misuse. The second requirement would be easily proved with a voice clone, but in order to meet the other two requirements it is likely that one would have to be a celebrity, I don’t think that I have enough of a reputation, and a simple voice clone would be unlikely to produce damages to an individual member of the public.

It’s therefore unsurprising that most applications of passing off to protect individuals have been celebrities (see the famous Rihanna passing off case).

Concluding

Protection against voice cloning may be dependent on a variety of factors. Privacy and data protection may still be the best avenue to stop such cloning, this is because at least in the UK IP protection, specially copyright, may be hit and miss and may also depend entirely on a number of conditions being right in order to gain protection.

The reality is that we could use with a much stronger protection for image and personality rights in the UK, perhaps in line with protection for likeness and voice in US publicity rights.

While we wait for reform, I’ll just be over here pondering the sad fact that my voice clone is a better public speaker than me.