Cartesia voices speak up to 25 languages: voice plus language on Sume

Cartesia says more than 50 of its library voices speak up to 25 languages natively, checked against a native-speaker rubric. On Sume you pick a voice id and set a language per job, and the platform warns when the pair looks mismatched, so a multilingual voice means you test each language you need rather than trusting a count.

Treat the vendor's 25 as an upper bound for the best voices and as a prompt to audition.

What Cartesia says

The Cartesia multilingual voices post (2026-09-23) describes 50+ library voices speaking up to 25 languages natively, voice cloning from ten seconds of audio, and a native-speaker rubric used to judge them. The wording is "up to": not every voice covers all 25.

What the Sume request needs

The TTS contract takes a voice from avatar_id, avatar_handle or voice.id (a UUID or a voi_ library id), and a language as a BCP-47 code of 2 to 16 characters. Omitting the language defaults to English at the provider, with a fallback that infers Korean or Japanese from a Hangul-only or kana-only transcript. Never rely on that fallback for other languages; set the language every time.

When the voice and language look inconsistent, the request returns a warning and you resend with confirm_language_mismatch set to true after a person has agreed. The confirmation does not change the voice or the language you asked for.

A matrix test

List your target languages in rows and your candidate voices in columns. Generate the same 15-second script, translated, for each cell and mark the ones you would ship. Do not extend a passing result to a neighbouring language; accent and pacing change between them. The 23-language audition post has a ready-made layout.

• Use a script with numbers and a proper name in every language.
• Have a native speaker listen, as Cartesia's own rubric does.
• Record voice id, language and model id from the job result.
• Re-test after any model change.

Cloned voices

Cartesia's cloning from ten seconds is a Cartesia product feature. Sume documents no public cloning endpoint, as the earlier voice cloning post explains, so plan around the voices you can select today.