Thanks — that sounds frustrating. I’m responding to your issue about Deepgram randomly switching languages while you’re speaking English. What to try first (most likely fixes) • Lock the language: Set the

language

parameter to

en

(or

en-US

) in your STT request instead of using

multi

. Multilingual/auto-detect mode will actively switch languages and can mis-detect when audio is noisy or non-standard. • Use a high-accuracy English model: Select an English-optimized model (for example,

deepgram/nova-3-general

) with

language="en"

for more stable results. • Improve audio quality: Reduce background noise, ensure clear mic placement, and check levels. Poor audio or overlapping speakers can trigger false detections. If you already lock the language and still see switches • Confirm the API request being sent actually includes

language: "en"

(copy the request payload if possible). • Test with a short, clean English clip to rule out environment issues. • Try a different English model to see if the behavior persists. If you’d like, I can help further — please tell me: 1. Are you currently sending

language: "multi"

or a specific language in your requests? 2. Which Deepgram model are you using? 3. Is this happening on live streams, recorded files, or both? Once you share those details I’ll give targeted next steps you can apply right away. Sources: Deepgram STT | LiveKit Documentation | Troubleshooting STT not picking up utterances | LiveKit