Deepgram Aura-2 languages: seven, and how Sume TTS sets a language

Deepgram's TTS models page lists seven Aura languages: English, Spanish, German, French, Dutch, Italian and Japanese. Sume TTS does not publish a fixed language list in its request schema. It takes a language field (2 to 16 characters, BCP-47 or ISO-639) that must match the voice you picked, and warns before a charge if it does not.

Deepgram's list is from its models page, read 2026-10-01; Sume's from the OpenAPI schema behind the API reference.

How many voices does each Aura language have?

The page gives a voice count per language.

How does Sume decide which language is spoken?

From the language field. The schema says to set it for every non-English transcript: if omitted it defaults to English at the provider, though Sume infers Korean or Japanese from a transcript that is only Hangul or kana. It also says never to translate a non-English request into English.

What happens if the voice and language disagree?

A known mismatch returns 409 tts_voice_language_mismatch before any job or charge. Retry with confirm_language_mismatch: true only after you have checked the voice, transcript and language. Regional tags compare by primary language, so en-GB against an English voice does not trip it.

Which one should I choose for Japanese or Dutch?

Both appear workable on paper, but a language list is not an audio test. Deepgram names Dutch and Japanese; on Sume, pass the language and listen to the result. For the Sonic side of that question, see Sonic 3.6 languages.