No language warning on Sume TTS? Raw voice ids skip the check

Sume TTS 1.0 stops a Korean voice reading a Japanese script, but only when it knows what language the voice speaks. If library metadata is unknown or unavailable you can still submit a raw voice id. In that case there is nothing to compare, so no 409 and no warning.

What counts as a raw voice id

The voice.id field takes a TTS voice UUID (8-4-4-4-12 hex) or a Voices library id (voi_ plus 32 hex). Anything else, such as a voice name from another vendor, is rejected at once with 400 invalid_voice_id before a job is queued or credits are reserved. A UUID that Sume has no language record for passes shape validation, and nothing is compared.

Why it matters

The mistake this check exists for is quiet. A voice reading text in a language it does not speak produces audio, and you pay for it: at $0.0475 per 1,000 characters, a 20,000-character script is $0.95. Nothing errors. You find out when you listen.

Build your own safety net

• Keep a small table in your project: voice id, language, region, who approved it.
• Send the language field on every non-English request, so the request states the language you intend.
• Render the first 200 characters of any new voice and script pair as a 1 cent test before the full job.
• Prefer an avatar or library voice, where Sume knows the language, over a bare UUID.

A cheap test

Submit the same short script with the new raw id and with a library voice in the same language. If the raw id sounds off, you have spent about 2 cents. If you skip the test and the full script is wrong, you have spent up to 95 cents per 20,000 characters, plus any video render that used the audio.