Ukrainian (uk) language support for Flux (flux-general-multi) #1596
Unanswered
admin-opv
asked this question in
Feature Requests
Replies: 3 comments 1 reply
|
Thanks for asking your question. Please be sure to reply with as much detail as possible so the community can assist you efficiently. |
0 replies
|
Hey there! It looks like you haven't connected your GitHub account to your Deepgram account. You can do this at https://community.deepgram.com - being verified through this process will allow our team to help you in a much more streamlined fashion. |
0 replies
|
It looks like we're missing some important information to help debug your issue. Would you mind providing us with the following details in a reply?
|
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi Deepgram team,
I'm building a voice AI product (a real-time conversational agent) and am currently using flux-general-multi for STT in production. Flux's turn detection is exceptional — the ~260ms end-of-turn latency and EagerEndOfTurn/TurnResumed events make it far superior to Nova-3 for our voice agent use case.
However, we need Ukrainian language support, which Flux currently lacks.
The problem:
We serve Ukrainian-speaking users. Nova-3 supports Ukrainian (uk) and transcribes it well, but its silence-based endpointing doesn't match Flux's conversational turn detection quality — which is critical for our real-time voice interaction flow (barge-in interruption, natural turn-taking).
Our current workaround is passing language_hint=ru when the user selects Ukrainian. Flux does produce somewhat intelligible output since the languages share phonetics, but the transcripts come out as a Russian/Ukrainian mix ("surzhyk"). This breaks our downstream RAG pipeline. The knowledge base is indexed in Ukrainian, and the garbled transcripts don't match, which significantly degrades retrieval relevance and response quality.
Why this matters:
Ukraine has ~45M native speakers and a growing demand for AI products since 2022
Deepgram already has Ukrainian acoustic/language models for Nova-3 — the foundation exists
For voice agent builders using Flux, the absence of uk means choosing between great turn detection (Flux) and correct transcription (Nova-3) — we shouldn't have to make that tradeoff
All reactions