Sume TTS language field: set ja or ko, or the voice reads in English

Set language on every non-English Sume TTS request. If you omit it, the provider defaults to English; Sume infers ko or ja only as a fallback when the transcript is all Hangul or all kana. A mixed script, such as Japanese text with a Latin brand name, is not covered by that fallback.

The field takes a BCP-47 or ISO-639 code of 2 to 16 characters, such as ko, ja or en. This is from the TTS request schema in the repository (read 2026-10-09).

What the schema says

The schema's guidance is blunt: never translate a non-English request into English, and set the language for every non-English transcript. A separate flag, confirm_language_mismatch, exists for the case where the voice and the language do not match. Send it only after the user confirms the warning, and omit it on the first request.

Why a default of English is expensive

A wrong-language read is billed like a right one: the charge is per character, so a 5,000-character Japanese script read as English costs 5 x $0.0475 = $0.2375 and gives you nothing to use. Multiply that by a batch of 200 scripts and you have paid $47.50 for audio you must regenerate.

The cheap defense is a one-line test per language and a request builder that refuses to send a non-English transcript without language. Make the field required in your own code even though the API treats it as optional.

A Korean example with the price

A 600-character Korean product line costs 0.6 x $0.0475 = $0.0285, and characters count the same way as Latin ones, including spaces and punctuation. The request is the same as an English one plus "language": "ko".

curl -X POST https://api.sume.com/v1/tts-1.0/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ko-line-001" \
  -d '{"transcript": "안녕하세요. 오늘의 신제품을 소개합니다.",
       "avatar_handle": "studio_presenter", "language": "ko"}'

Captions follow the language too

If the audio goes under a video that you caption, the video captions job needs a Hangul style for Korean speech. slam, punch and tiktok-green use Latin faces and reject Hangul text with caption_hangul_text_latin_style. The language field on the caption job is only a speech-to-text hint and never picks the style.

A pre-flight checklist

Before a multilingual batch, check four things: every request carries language, the voice you picked speaks that language, mixed-script lines have been tested on their own, and the character counts include spaces and punctuation. A one-line test per language costs about a cent.

If the voice-language mismatch warning appears, do not set confirm_language_mismatch blindly. Ask whether the voice or the language is wrong, since confirming does not change the requested voice or language.

The same applies to the TTS Router: its request is TTS 1.0's plus a required model, and language works the same way there. Both are billed at $0.0475 per 1,000 characters.