AI Voice Contracts for Voiceover Artists: Why the Industry Still Has a Long Way to Go
Updated: 11 hours ago
Text to speech technology goes back further that you might think. Way further. In 1791 as it happens, Wolfgang von Kempelen built the first mechanical speaking machine using bellows, reeds, and resonating chambers to form simple words and phrases. In 1846, Joseph Faber unveiled the "Euphonia," a keyboard-operated mechanical speech machine with rubber and metal lips and a tongue. Cut to 1938 and Homer Dudley at Bell Labs invented the Voder (Voice Operating Demonstrator), the first electronic speech synthesizer, famously showcased at the 1939 New York World's Fair.
Closer to own time, text to speech (TTS) jobs for VOs used to be the stuff of train station announcements. Hundreds of thousands of different words and phrases recorded by an ever calcifying voice artist over what was mostly likely thousands of hours of studio time.
Today, TTS is one of a suite of VO "jobs" offered by the ever expanding synthetic voice industry. If you're a VO you'll recognise these job briefs, they use words like "cloning" and "data sets." They'll often appear with seemingly humungous fees attached. "$35,000 for 100,000 words" was one I saw recently. Impressive though this sounds, is it actually enough to compensate an artist for what they're providing?
A little while ago I was sent a contract for what was described as a text-to-speech recording project. Ten hours of recording, a three-year licence, and a headline rate that looked fabulous by normal rate standards. Then I read the paperwork and turned it down. After a bit more digging, I discovered that this was a fairly typical example of where AI voice contracts are right now: written almost entirely for the benefit of the company, and handed to voice artists as though that’s standard practice.
It isn’t standard practice. Or at least, it shouldn’t be. This post is about what I've found out about rates for AI voiceover work, what the unions and industry bodies are doing about it, and what a good AI voice agreement should actually look like.
What was in the contract
So, to use my recent experience as an example: I won’t name the company, but I will name the platform, (voices.com ) and l'll describe the terms:
Everything was assigned, not licensed. Ownership of the recordings, the resulting voice model and anything built from it passed to the company from the moment of creation, along with a waiver of any voice, personality or portrait-right claims.
The permitted uses went far beyond the project. The brief described a chatbot voice. The contract allowed digital humans, advertising, broadcasting, audiobooks, customer service and dubbing, which is a fairly complete list of the markets I work in.
Sublicensing was wide open. Affiliates, business partners, enterprise customers and end users could all be given rights.
The three-year stated term didn’t really end anything. On expiry the company might delete the source recordings, but any model or product already built from them was carved out. You get paid for three years while the clone carries on.
There was no fee in the document. The contract said no additional payment was owed for any use. Whatever I was paid would have to sit in a separate agreement, while the rights were assigned directly to the end company.
Everything sat in someone else’s jurisdiction. The governing law and courts were in a country I don’t live or work in, and a non-disparagement clause covered the company’s products.
None of that was hidden. It was all there on the page. It just doesn’t match the way the work was described in the brief.
Why these contracts are so lopsided
From what I can tell, a few things are going on at once.
Old contracts are being stretched to cover new technology. Equity, the UK performers’ union, has said it is exploitative to build digital replicas from recordings made under contracts that pre-date generative AI or were never drafted with it in mind. Voice artists in Scotland have gone public after finding synthetic versions of their voices in (you guessed it) public transport announcements, built from recordings they made years earlier for other purposes.
“Research” and data collection jobs aren’t always what they seem. Equity has described artists taking work to record their voice for research with AI developers, only to find they had assigned their rights to an end client who could use the voice, or a synthetic copy of it, commercially.
The law hasn’t caught up. The IP lawyer who helped develop Equity’s toolkit has called it a very good but temporary fix until the government reforms the law, and has said the UK’s legal framework isn’t well designed to protect performers from unauthorised imitations using AI. Protections vary a lot from country to country, so for now much of the protection has to come from the contract itself.
Buyers hold the leverage. Intermediaries often invite your ‘most competitive rate’ because their client is comparing several suppliers. A generous-looking hourly figure is easy to say yes to. The rights grab is on page three.
What the unions and industry bodies are doing
None of these are law, but they give you something concrete to point to when you ask for changes.
Equity (UK) has published an AI toolkit for members, with model clauses to stop performances being cloned without consent, a template AI contract, a template take-down notice and a vision statement. The vision statement says artists should have consent for past, current and future performances, fair and proportionate remuneration, and engagement under collectively bargained agreements.
SAG-AFTRA (US) has been writing AI terms into its contracts. Its 2025 video game agreement requires informed consent for every way AI is used on a performer’s voice or likeness, sets minimum rates for digital replicas, and pays at least 7.5 times the minimum when a replica is used for real-time generation, such as a chatbot. Its TV animation contracts state that a voice actor is a human, and require producers to get consent before using a performer’s name to prompt a synthetic voice.
NAVA (US) publishes an AI and synthetic voice rider that anyone can add to a contract. Its principles include the ability to consent to use, explicit limits on use of the results and proceeds, and the ability to opt out or set term limits.
The Association of German Voice Actors publishes contract conditions that include written consent to any new use, enough time to have the contract reviewed by a lawyer, the contract in the artist’s own language, and jurisdiction in the artist’s own country.
MEAA (Australia) says in its rate card that voices created with generative AI should be paid at the same rates and for the same usage periods as a studio-recorded voice, and that using recordings to train AI is prohibited without the artist’s explicit and informed written consent.
So, what does a GOOD AI voice agreement look like?
Take the common threads from all of the above and I at least have concluded that a good agreement has these values at its core.
Informed, specific consent. You know exactly what is being built, and for whom, before you sign. Consent for one product isn’t consent for everything.
A defined use. The named product or client, the territory and the platforms are written down. Anything new needs your written agreement first.
A licence, not an assignment. You keep ownership of your voice. The company gets permission to use it for a stated purpose.
A real end date. When the term ends, the model is retired or deleted, not just the source files. Renewal is a new conversation with a new fee.
Fair, separate payment. Recording time and usage are paid for separately, and usage reflects scope, territory and duration, as it does for any other voice work. Where the use is ongoing, there should be reporting, ideally with a right to audit.
Real limits. No use for political, adult, misleading or defamatory content, no sublicensing to third parties without your approval, and no creating a public voice library or voice product from your recordings.
A fair process. The contract is in your language, governed by the law where you live, and you have time to get it checked before you sign.
Red flags?
A broad assignment of rights instead of a licence
Waivers of voice, personality or moral rights
A term that expires on paper but not for the model
No fee written into the contract itself
Free auditions that run to several minutes of natural, unscripted speech, which is voice data
Pressure to quote your most competitive rate against unnamed rivals
A contracting company that’s different from the one in the email
What I do
When an AI voice enquiry arrives, I ask the same questions every time. Who is the end client? What exactly will the voice be used for? For how long, and what happens to the model afterwards? Is it a licence or an assignment? Where does the money sit? If the answers are clear and the terms are reasonable, I’m happy to talk about price, and I treat the usage like any other: by scope, territory and duration. If the answers are vague, or the contract is built to take everything, I decline politely and say why. Some of the better enquiries I’ve seen arrive with protections already built in, which tells me it can be done.
If you’re a company building voice AI
The shortcut is tempting, but voice artists talk to each other. Clear, proportionate terms are the easiest way to attract good talent and keep it. Use the Equity and NAVA materials as a starting point, state your use plainly, pay properly, and give people time to read.
The bottom line
AI voice work isn’t going away, and despite it giving me the ick, I’m not completely against it in principle. But the agreements still lag far behind the technology, and until the law catches up, the contract is the only protection a voice artist has. Read every clause. Ask the awkward questions. And if a contract takes everything and doesn’t even say what it pays, the answer is no.
If you’ve been sent an AI voice contract and want a second opinion on the rate, you’re welcome to get in touch.
Melissa Chambers is a British and Australian voiceover artist with 20 years of experience, working from a professional home studio in the UK. You can listen to her demos and get in touch at melissachambersvoice.com.
Comments