Gradium TTS covers 5 languages: what Sume's language field does

On Sume TTS you set the spoken language with the language field; leaving it out means English. Gradium's October 7 voice update covers five languages (English, French, Spanish, German, Portuguese), and the same five can be requested on Sume only as far as the Sonic engine and the voice you pick support them, which the schema does not list per language.

So the honest answer to "does Sume cover Gradium's five languages" is: you can submit them, you cannot assume every voice is fluent in each, and Sume has a guard for the case where the voice and language disagree. Sources: Gradium's announcement (read 2026-10-10) and the Sume schema in the API reference.

What does Gradium's page claim?

Per the page, the catalog now has 28 accents across five languages. English has 10, Spanish 7, French 4, German 4 and Portuguese 3. Gradium frames accents as part of the voice, so a customer chooses a regional voice rather than a language switch.

What does Sume's language field do?

In the Sume schema, language is a 2 to 16 character BCP-47 or ISO-639 code such as en or ko. The description says to set it for every non-English transcript, because an omitted value defaults to English at the provider. It also says Sume infers Korean or Japanese from a transcript made only of Hangul or kana as a fallback, and never translates a non-English request into English.

For the five Gradium languages the codes are en, fr, es, de and pt. Sume does not translate: a French script sent with no language is read by an English-defaulting call, which is a common cause of an odd accent. Send the code every time.

• Set language on every request that is not English.
• Write the transcript in the target language yourself; Sume TTS does not translate.
• Regional accent is a property of the voice you choose, not of language.

What is the language mismatch guard?

When the chosen voice and language disagree, the API warns instead of silently producing a wrong accent. You resend the same request with confirm_language_mismatch: true only after a person has accepted the warning. The schema states that confirmation does not change the requested voice or language, so it is a consent flag, not a fix.

What should you test before committing?

Pick one line per language, run each through the voice you intend to use, and listen. Cost is small: at $0.0475 per 1,000 characters, a 200-character test line is 0.95 of a cent before rounding, and Sume rounds each job up to a whole cent, so five test lines are five cents. Keep the passing combination (voice id, language, engine id) in your config so later episodes repeat it.