Voice Library Licensing: Three Permissions, One Checkout

A voice library licence has three layers: the right to use the catalog, the right to use the output, and the right to use a person’s voice. Most checkout pages document only the first, and many do not document even that. The gap is not a loophole; it is a procurement risk that shows up when a client asks for proof.

This guide covers what the ten providers in our 2026 ranking state about each layer. It is not legal advice. It is a map of the public record.

Layer one: the catalog

A catalog licence says the voices in the library are cleared for use by the vendor. Deepdub’s “1,000+ licensed voices” is the clearest case here: the word “licensed” is attached to the catalog itself.

A catalog licence differs from a commercial output licence. It tells you the vendor has the right to offer the voices; it says nothing about what your plan allows you to do with the audio.

Layer two: the output

The output licence is the line that says “Commercial License” or “commercial use” next to a plan. Only one provider in this run states it clearly on the page where the catalog is sold: ElevenLabs attaches a Commercial License to Starter.

The others are silent in different ways. Google Cloud TTS prices characters and requires billing but does not carry a commercial-licence line. Amazon Polly does the same. Speechify sells a consumer reading plan. Murf’s plan page did not render for us. Kits AI, Respeecher, HeyGen, and Synthesia do not state output rights on the pages we checked.

Silence is not prohibition, and it is not permission. It means the buyer has to ask.

Layer three: the voice identity

If the voice is a stock catalog entry, the vendor’s licence usually covers the catalog voice. If the voice is a clone of a real person, the person’s consent sits above the platform licence. No platform can grant a right it does not hold.

Cloning belongs in four situations:

1. your own voice;
2. a voice actor who has given written consent for the project and territory;
3. speech restoration for a person with a speech or language condition, with consent;
4. post-production on a signed performer whose contract covers synthetic replacement.

Outside those four, the honest answer is a lawyer and a signed agreement, not a tutorial. This site does not publish instructions for cloning someone else’s voice.

The provider table

The pattern is stark. The catalog column is full; the output column is nearly empty. That is the finding, and it is more useful than a ranking score.

What to ask before you pay

• Which plan carries the commercial right, and what does it exclude?
• Does the right cover client work, paid social, broadcast, and resale?
• If the voice is cloned, who holds the consent, and can I see it?
• What happens to the licence if I cancel?
• Can I get the terms in a document, not just a marketing page?

If a provider cannot answer the first two, the subscription is a test, not a clearance.

Why this matters for a catalog ranking

Our method gives licensing clarity 20% of the weight because it is the field most likely to stop a delivery. A library with 10,000 voices and no licence line is harder to use than a smaller library with a clear one. The ranking rewards disclosure, not catalog size alone.

For the full order, see the 2026 results.