Deepgram Nova-3 de-CH, Lithuanian, Portuguese: language_code on Sume
Deepgram improved Nova-3 for Swiss German, Lithuanian and Portuguese on September 24. What to send as language_code on Sume STT for regional variants.
Deepgram's September 24, 2026 changelog lists improved Nova-3 monolingual models for German (Switzerland) de-CH, Portuguese (pt, pt-BR, pt-PT), Lithuanian lt, and, for streaming only, Flemish nl-BE. On Sume STT the equivalent decision is the optional language_code hint: send a BCP-47 style tag, or omit it for auto-detect.
Sources: the Deepgram changelog and Sume's API reference, read 2026-10-01.
Which models did Deepgram update?
The page says no request changes are needed: the updates are live for all users.
| Language | Codes | Where |
|---|---|---|
| German (Switzerland) | de-CH | Batch and streaming |
| Portuguese | pt, pt-BR, pt-PT | Batch and streaming |
| Lithuanian | lt | Batch and streaming |
| Flemish | nl-BE | Streaming |
What do I send to Sume for Swiss German or Brazilian Portuguese?
The Sume schema calls language_code an optional BCP-47 or provider language hint, with en and ko as examples, and says to omit it for auto-detect. The video inspect docs repeat that it is an STT hint. Neither source lists which regional tags change the result, so do not assume de-CH behaves differently from de.
How do I check a regional variant?
Run the same 30-second clip three ways: with the regional tag, with the base language tag, and with no hint. Read the three transcripts for the words that differ by region, such as vocabulary and spelling, and keep the setting that reads best. Completed results include words[] timings and language fields when available, which helps you spot a clip detected as the wrong language.
Can I choose the Nova-3 model on Sume?
No. Sume exposes sume/stt-1.0 and keeps provider models internal, so a vendor's model update shows up as a quality change, not a parameter change. For another language-code case see Kazakh language codes.
Sources
Related posts
More in Models
- ElevenLabs character limits by model vs Sume TTS 20,000
ElevenLabs lists 5,000 characters for v3, 10,000 for v4 and 40,000 for Flash v2.5. Sume TTS 1.0 takes up to 20,000 characters in one request.
- ElevenLabs speech to speech API: what Sume offers instead
ElevenLabs lists speech-to-speech voice changer models. Sume has no voice-to-voice endpoint: transcribe with STT, edit the text, then run TTS.
- ElevenLabs STT edit instructions vs Sume transcribe: what Sume returns
ElevenLabs added transcript-editing instructions up to 2,000 characters to its speech to text. Sume's transcribe endpoint returns a plain transcript.
- ElevenLabs dialogue continuity vs Sume TTS segments: what differs
ElevenLabs Text to Dialogue links takes with previous_text and request ids. Sume TTS has no such fields; it returns gapless sentence segments instead.
Written by Sume