Sume Voices page lists 16 languages, Sonic 44: which list applies?

The two lists answer different questions. The Voices page in the Sume web app offers 16 languages when you create a voice; the Sonic page from Cartesia states 44 languages for the speech model; and the TTS request just takes a BCP-47 language code that you set per job. Check all three if you create a voice and then synthesize in a language outside the 16.

The safest rule is to create the voice in the language you will use most, and audition it in every other language before you ship.

The three places a language shows up

In the Voices page, a language selector lists English, Korean, Japanese, Chinese, Spanish, French, German, Portuguese, Italian, Hindi, Dutch, Polish, Russian, Swedish, Turkish and Tagalog. In the TTS contract, language is a string of 2 to 16 characters; non-English requests must set it, and omitting it defaults to English at the provider. On the Cartesia Sonic page, the model is described as supporting 44 languages.

What can go wrong

If you create a voice in English and ask it to read Korean, the request can return a mismatch warning. You can confirm it with confirm_language_mismatch, but the confirmation only acknowledges the warning; it does not change the voice or the language. Hearing the result is the only check.

• Create the voice in its main language.
• Set language on every non-English request.
• Audition each extra language with a native listener.
• Keep the model id fixed while testing.

A practical order

Decide the languages you need, check them against Cartesia's list, create or choose a voice in the closest of the 16 Voices languages, and then run a short audition matrix. If a language you need is outside the 16, test whether an existing voice speaks it acceptably rather than assuming it will not. The audition matrix post gives a layout.