By Vocal Variants

It seems like with each week comes a new article about a miraculous new artificial intelligence (AI) algorithm that sounds eerily close to a real human voice, spawning debate and panic in various voice over communities.

But we're not here to argue the quality of these synthetic voices, or the timeline of its proliferation - because like it or not, AI voices are already here.

Trying to stop it is like putting your body in front of a moving train. The best you can hope for is slowing it down.

Instead, with the advent of any new technology, performers can be a part of setting the tracks to guide which path synthetic voices (yes that's the train in this analogy) take in the future.

First things first, what do we mean when we say synthetic or AI voice?

AI stands for artificial intelligence. It is an area of computer science that uses computer programs to imitate human intelligence. AI can have many implementations outside of creating a synthetic voice.

 A synthetic voice is a voice that is produced by machines instead of a person. In order to train AI how to simulate human voice, it needs data to study (our voices). This is called machine-learning.

Once a program or algorithm can satisfactorily replicate human sounds, it can create a number of different sounding synthetic voices.  

A synthetic voice, or digital voiceprint, can be based on:

There are two main methods of creating synthetic voices: text-to-speech and speech-to-speech.

Text-to-speech is when the user inputs words (text) into the algorithm and the output is the speech, the digital voice "saying" the words inputted (speech). This is what is most commonly referenced when discussing AI, as the text-to-speech technology is closer to creating sounds that mimic human speech than its counterpart.

This brings us to speech-to-speech. Instead of inputting text into the algorithm, one inputs, you guessed it, speech (or audio). A director could essentially input a line read, or a performance in a different language, into the algorithm to dictate the pacing and inflection for the synthetic voice.  

With all this new innovation, it's crucial that all voice artists, union and non-union, know how to protect their vocal identities.

How do we do that? The key is contracts.

Nobody loves reading contracts, but now more than ever, it is important that we do. And carefully. When it comes to protecting our vocal identities, we need to look at the contracts for BOTH non-AI related jobs, and AI jobs.  

For now, let's discuss contracts for the first type of job, the non-AI related ones.

While it seems like all these AI voice algorithms and technology have popped up overnight, what is increasingly clear is that they were quietly being developed for quite some time. Stories and rumors are coming to light about pay-to-play sites selling audition files to AI devs, or databases, for AI voice apps that were initially said to be used for charity, but then sold to for-profit companies marketing to advertising agencies.

Our vocal data has been used, maybe even legally, and we didn't even know it. Damn. We should've read that contract.  

At this point, a decent number of AI algorithms have already been developed and built (the train's already left the station, if we're keeping with our analogy). But moving forward, if an actor doesn't want their voice to be used in further development, what should they look out for in their contracts?

If you are hired on a non-AI related job, and you want to make sure what was recorded isn't also used or sold to develop a synthetic voice, or train an AI algorithm, you need to pay attention to the Usage section of your contracts. This section defines and describes how your voice will be used. It can also be called "Terms of Usage," "Rights," "Assignment of Rights," "Licensing," or a variation of the above.

Most contracts will not spell out that they are reserving the right to sell or use your vocal data for AI purposes (that would set off too many alarm bells).

Instead, vague and over-arching language is used to encompass that usage.

Actors should be wary of any broad language that gives away your rights "in perpetuity" or "throughout the known universe." Other things to be on the lookout for are clauses that grant the producer:

These types of rights mean the producer retains the right to use your voice in any manner of future use (that may have not even been invented yet), without needing to renegotiate or compensate you.

In general, the broader the usage, the higher the likelihood the client can use your vocal data for things outside of the project you are recording for, i.e. training or creating synthetic voices. The narrower the usage, the more one is protected.