STT issue for Armenian language or voice

What do you need help with?

Feature request

Issue area

Speech Recognition / STT

Impact

Blocker - I cannot continue

Character ID

aa5b2858-6eef-11f1-b52e-42010a7be02e

Session ID

559601a2b42f56faf95a073a263abd1e

Where does it happen?

Web SDK / Embed

Language

Armenia

Voice provider

Azure

Voice name

Andrew Male Multilingual Voice

Short summary

Avatar cannot understand Armenian language voice input

User input

one of our characters has English and Armenian set as languages. typed Armenian input works fine end to end, character responds in Armenian text and speaks it back correctly. but when we speak Armenian out loud into the avatar, it’s not being transcribed correctly at all, so the character can’t respond properly to spoken Armenian. English voice input works fine on the same character, it’s just Armenian speech that’s failing. our account only has Deepgram available as the Speech Understanding provider for this character, no option to switch to anything else. is there a way to get a different STT provider for this language, or any other fix on your end? typed input works so it seems isolated to Deepgram’s Armenian speech recognition specifically.

What happened?

We tried to speak in Armenian or generated an audio speaking in Armenian language but it is not being captured by the avatar and still looks for English understanding and words

What did you expect?

Its should capture Armenian speaking similar to how it captures English speaking

I checked the relevant documentation, resources, and similar forum posts before posting.

on

I confirm that I did not include API keys, tokens, passwords, secrets, private character data, private conversation logs, or customer data.

on

Hi Martin, thanks for the detailed report. We can see this session reached a legacy realtime path, so we want to confirm exactly which client and stream type you are using before drawing a conclusion about Armenian STT.

Could you please share:

  1. The exact page or application where you are testing: Avatar Studio, a published Convai Experience / Embed, a direct Pixel Streaming URL, or a custom Unreal integration.
  2. If embedded, the Embed package/version you use, or whether you are embedding the Pixel Streaming iframe directly.
  3. Your browser, operating system, and microphone device.
  4. How voice input is started and stopped: the T key, an on-screen button, or a custom integration.
  5. A new session ID and approximate UTC time immediately after reproducing with a short Armenian recording. A screen recording would also help.
  6. Whether the same character and Armenian audio work differently in Avatar Studio versus the embedded experience.