AI Voice Licensing Rates: What Platforms Actually Pay

Two numbers decide what a synthetic voice is worth right now, and they are nowhere near each other. ElevenLabs says voice creators on its marketplace have earned over $22 million in total. The European speaker associations put the minimum fee for building one synthetic voice at €5,000 — before a single second of that voice is used commercially.

Both figures are published. Neither is secret, and neither is disputed. They are almost never printed side by side, which is why voice actors keep signing terms they have not compared to anything. This page collects what the platforms and the associations actually publish, with the source behind every figure.

Three different ways a voice gets paid for

Almost every offer a voice actor receives in 2026 is one of three structures, and the words used to describe them are often interchangeable when the economics are not.

• Usage royalty. The platform hosts a clone of your voice and pays you a micro-amount each time a paying customer generates audio with it. Income is open-ended and unpredictable. This is the ElevenLabs Voice Library model.
• One-off data sale. You deliver recordings once, get paid once, and the buyer trains a model with them. No clone is offered back to the market under your name. Murf runs this as a separate programme.
• Licence fee per use. The classic voice-over model: a fee for the session, a separate fee for each defined use. The European associations argue this logic does not change just because the audio was generated rather than recorded.

The confusion that costs money is treating the first as a version of the third. A royalty rate per 1,000 characters is not a licence fee — it does not distinguish a regional radio spot from a national TV campaign, and it does not scale with reach.

What ElevenLabs publishes about payouts

ElevenLabs is the only large platform that publishes hard aggregate numbers, so it is the one that can be checked. In a company post dated 22 May 2026, it reported that voice creators had earned over $22 million in total, up from $11 million in November 2025, across more than 10,400 creators and voices in 32 languages.

Divide one by the other and the lifetime average per creator is roughly $2,100. That average says very little on its own: the same post profiles creators who earn a full-time living from a single voice, which means the distribution is heavily skewed and most participants sit well below the mean. Treat marketplace income as a lottery with a long tail, not as a salary.

The mechanics are documented in the ElevenLabs help centre and product docs:

• Only a Professional Voice Clone can be shared. Instant clones and designed voices cannot.
• Rewards accrue only on generations by paying users. Free-tier usage of your voice earns you nothing.
• Your default rate is set by the notice period you choose — the minimum is three months, the maximum two years, and a longer period pays more.
• Earnings from the ElevenAgents product are capped at $0.01 per minute of conversation.
• Payouts run through Stripe Connect roughly weekly above a $10 threshold, require an active paid subscription, and are available only in the countries Stripe Connect supports.
• Because ElevenLabs is a US company, payouts require W-8 or W-9 certification. Without a treaty claim, the default withholding on earnings from US customer usage is 30%.

The clauses behind the payout

The commercial terms sit in the Voice Library Addendum, last updated 6 March 2026. Four provisions are worth reading before the rate card.

Outputs survive removal. Audio generated with your voice before the end of the notice period continues to exist and remains usable afterwards. Withdrawing your voice stops future generations; it does not recall the ones already made.

The notice period is a one-way door. Once set, you can increase it but not reduce it. A two-year notice period earns more today and keeps your voice in other people’s accounts for two years after you decide to leave.

Participation is discretionary. The Addendum states that participation in Financial Rewards is at the platform’s sole discretion and can be suspended or terminated, and that voices may be reviewed, renamed or removed without notice.

The royalty discharge clause. By sharing a voice, the addendum states that ElevenLabs and all other users are discharged, to the fullest extent permitted by law, from any present or future obligation to pay royalties or equitable remuneration for that voice. Whether such a clause binds a performer resident in the EU is a separate question — German law, for instance, provides in § 32a(3) UrhG that claims to further participation cannot be waived in advance, and § 79(2a) UrhG extends §§ 32 to 32b to performers. The clause is drafted to yield wherever local law says otherwise; nobody has yet litigated where that line falls.

One eligibility rule is telling on its own: residents of Illinois may not share a voice at all, and must remove it if they move there. ElevenLabs gives no reason. Illinois is the state with the strictest biometric privacy statute in the US.

What European associations say a synthetic voice should cost

The counterweight to the platform rate cards is the KI-Gagenkompass, published jointly by the German association VDS, VOICE in Austria, VPS|ASP in Switzerland and United Voice Artists, last updated in April 2025. It is guidance rather than a tariff, but it is the most detailed public benchmark that exists for synthetic voice work in Europe.

• Voice synthesis session: €1,000 / €1,250 / €1,500 per recording day (lower, middle and upper average), one day meaning a maximum of five hours including breaks.
• Minimum fee for a base synthesis: €5,000 / €6,250 / €7,500, applied when fewer than five recording days are needed — for example when usable material already exists.
• Listing fee: from a symbolic €150 per year up to several thousand euros per year, payable for the right to offer the synthetic voice in a catalogue at all.
• Neural learning: no rate is published. The associations advise members not to participate, and state that if it is agreed anyway the value should be at least six figures.
• Exclusivity: also placed in the six-figure range.

The structural point matters more than the numbers. In that model, the synthesis fee, the listing fee and the usage licence are three separate payments, and none of them includes the others. The document is explicit that the synthesis fees grant no usage rights whatsoever. A marketplace royalty collapses all three into one per-character rate.

The associations also recommend contract terms that no self-serve platform currently offers: a defined end date, deletion of the voice data on request, a ban on blended or morphed voices, approval rights over output quality, and the performer’s country of residence as the place of jurisdiction. Our comparison of union and association rates in the US, UK and Germany covers the underlying licence logic in more detail.

One-off data deals are a different transaction entirely

Not every AI voice offer creates a clone. Murf runs a voice data sourcing programme described as a single transaction with instant payout and explicitly no clones: recordings are used to train base models, not to produce a voice sold under your name. The published requirements are at least 20 minutes of continuous audio at 44.1 kHz or above, minimal background noise, no overlapping voices, any language or accent, plus a voice-fingerprinting script for biometric verification. Compensation is stated to depend on language, quality and length; no figures are published.

Respeecher takes a third position: its voice actor pages state that talent set their own pricing and retain control over how the voice is used, with public and private models, but no rate card is published at all. When no numbers are public, the association benchmarks above are the only reference point a performer has going into the conversation.

Seven questions to ask before you list your voice

1. What exactly does the fee cover — the recording session, the right to list the voice, the usage, or all three bundled?
2. How do I withdraw the voice, how long is the notice period, and what happens to audio already generated?
3. Does free-tier or trial usage of my voice earn anything?
4. Can my voice be blended with others, retrained, or used to generate other languages?
5. Which content categories can I exclude, and is that enforced technically or only contractually?
6. Which law governs the contract, and in which country would I have to sue?
7. What is deducted before I see the money — withholding tax, payment processor fees, subscription requirements?

Question six is the one performers skip and lawyers start with. A €150 listing fee and a US venue clause are not the same deal as a €150 listing fee and a German one. If your voice is cloned without a contract at all, the applicable protections are set out in our guide to voice rights law country by country.

What this means if you are buying voice work

From the client side the arithmetic looks inverted. Marketplace generation is cheap per unit and comes with a commercial licence attached, which is why it has taken over first drafts, internal video and high-volume e-learning. What it does not give you is a named human being who can be re-briefed, a performance that was directed, or a rights position that survives a platform changing its terms.

For work where the voice carries the brand — campaigns, dubbing, anything a broadcaster will clear — the licence structure the associations describe is still the one that holds up. That is the model Voicfy is built on: quotes from native-language voice talent, with the usage defined before the recording rather than inferred from a character counter afterwards.

Both models will keep existing. The mistake is not choosing one — it is signing terms for the first while assuming you are being paid under the second.