The same voice, in a language it never recorded.

Dubbing usually means casting someone else. The voice that built the audience in one market is gone in the next, and what arrives instead is a competent stranger. Cloning inverts that: the identity is a request parameter, so the same speaker says the translated line.

languages from one cloned reference

Hear the identity survive the translation

Three takes below: one cloned voice, one sentence, three languages. Nothing was re-recorded and no second speaker was cast. The hard case in cloning is a language the reference never spoke, because the identity has to hold with none of the original phonemes to lean on — which is exactly what these three are.

One performance, three languages

Film

One take, one cloned identity, three markets. The dubbing arithmetic — and the same voice in English, Spanish and German — is below.

What this is, and what it is not

To be precise about the boundary: Gandr synthesises the dubbed audio and carries the identity across languages. It does not translate your script, separate a voice from a mixed soundtrack, or mux audio back onto video — those stay in your pipeline, where you already have tools you trust.

What changes is the part that used to require a casting session and a studio day per market: the voice itself, available on request, in every language you ship.

One reference clip, fingerprinted and cached — reuse across a whole catalogue skips re-cloning.
Per-request emotion and prosody, so a translated line can be delivered at the pacing the scene needs rather than the pacing the words happen to have.
Unmetered characters, which matters most here: a back catalogue is a large number of characters and a meter turns localisation into a per-title decision.

Notes — an engineer's checklist

01Does the cloned voice really hold up in another language?

That is the claim this page is built to let you check rather than take on trust: the three clips above are one cloned reference speaking English, Spanish and German, rendered on the production API. Published, dated specimens for eight languages live at /languages/.

02Do you translate the script?

No. Gandr synthesises speech and carries the voice identity; the translation stays with whatever you already use. Send us the translated text and the reference clip and you get the dubbed audio back.

03Can you separate a voice from a finished mix?

No. There is no source-separation or voice-isolation endpoint on this API, and we would rather say so than let you discover it mid-project. Bring a clean reference clip — five to ten seconds of one speaker is enough.

04What does a back catalogue cost to localise?

The seats you run it on, and nothing per character. That is the whole difference: on a meter, localising a catalogue is priced by its length, so the long tail never gets done.