One voice, many languages: Sume TTS 409 and confirm_language_mismatch
Microsoft's MAI-Voice-2.1 is reported to keep one voice across 23 languages. On Sume, a voice with a different language returns 409 until you confirm.
A third-party tracker reports that Microsoft's MAI-Voice-2.1, released 2026-10-01, covers 23 languages across 26 locales with one voice across languages (read 2026-10-05, tracker, not the vendor page). Sume's TTS has a different guard. Each voice has a primary language, and if your request names another language, Sume returns 409 tts_voice_language_mismatch before it creates a job or charges, until you confirm with confirm_language_mismatch.
How the guard works
Sume TTS reads the target from language on REST or payload.language on MCP. It compares that to the voice's primary language. A mismatch returns a 409 with voice_id, voice_language and request_language. On MCP, tts_create returns a non-error tts_voice_language_warning with confirmation_required: true.
| Case | Result |
|---|---|
| Languages match | No warning, no extra step |
| Known mismatch | 409 tts_voice_language_mismatch, no job, no charge |
| Retry after confirming | Same request and idempotency key, with confirm_language_mismatch: true |
| Regional tags | Compared by primary language; fil and tl match |
| Voice metadata unknown | A raw voice ID can still be submitted |
What to do
The repo docs say to ask the user, then retry the same request with the confirm flag, and not to change the voice, transcript or language. That makes the cross-language use a decision, not an accident. If you want a single voice across several languages, test each language with the confirm flag and listen, because the docs make no quality claim for a mismatch.
- Dry run first: a cost-only MCP
dry_runstays a preview, and the check occurs on submit. - Keep one voice id per language where you can.
- Do not read the MAI claim as a Sume feature. Sume's TTS router lists one engine, and the voice list comes from your library.
Sources
Related posts
More in Comparisons
- Unlimited plan break-even: Headliner $25.99, Zebracat $199 vs Sume
At what volume does an unlimited plan beat metered pricing? Headliner Pro and Zebracat's top plan against Sume's $0.10 render and $5.52 avatar ad.
- Veo 3.1 1080p needs 8 seconds: the short-clip trap and a 3 s option
Veo 3.1 requires 8 seconds for 1080p, 4K and reference images, so a 4 s 1080p clip is impossible. Omni on Sume does 3 s at 1080p for $0.5625.
- Veo 3.1 has no 1:1 square: 16:9 and 9:16 only, and where Sume does 1:1
Veo 3.1 supports only 16:9 and 9:16. Sume Omni is also 16:9 and 9:16 only; Seedance, Kling 3, Wan 3.0 and MiniMax H3 cover 1:1, and Seedance and H3 add 21:9.
- Veo 3.1 offers 4, 6 or 8 seconds; Sume Omni takes any 3 to 10
Veo 3.1 only accepts 4, 6 or 8 second clips. On Sume, gemini-omni-flash-1.1 takes any whole second from 3 to 10, and an 8 second 720p clip bills $1.00.
Written by Sume