Transcription not picking up new language
# support
b
Hey – am working on a multi-lingual assistant and can't get the transcriber to work in that language. My pipe: - TTS w/ Eleven Labs set to eleven_multilingual_v2 w/ a compatible voice - LLM is OpenAI 4o latest - STT (transcriber) is Deepgram Multilingual model Nova 2. I have the TTS working well – it can generate multilingual voice easily. However I'm running into an issue with the transcriber. I'm not able to get it to transcribe in the language that I want. The call flow starts in english asking which language the user prefers and then the assistant should pick up the language and go. What should I be changing to enable this? Without it, our service is pretty useless because we need to analyze the transcripts.
v
To enable automatic multilingual transcription with Deepgram Nova 2, you must set the
Language/Mode
to
Multi
.
For example, in your assistant configuration:
Copy code
yaml
transcriber:
  provider: deepgram
  model: nova-2
  language: Multi
This setting allows Deepgram to auto-detect and transcribe in the user's spoken language. Be sure to save your changes and test with different languages to confirm it works as expected ([see documentation](https://docs.vapi.ai/customization/multilingual)). Source: - [Multilingual Configuration Guide](https://docs.vapi.ai/customization/multilingual)
b
I already tried that – the transcripts look like gibberish and the model has no idea what I'm asking for. I have gotten it to work with Google but that wont' work because we need to be HIPAA compliant, and you only support AZURE and DEEPGRAM there.
c
Hi barelledboi, To enable automatic multilingual transcription with Deepgram Nova 2, make sure you have configured the transcriber with the
Language/Mode
set to
Multi
. Here’s how you can set it up in your assistant configuration:
Copy code
yaml
transcriber:
  provider: deepgram
  model: nova-2
  language: Multi
This configuration allows Deepgram to detect and transcribe the user's spoken language automatically. Ensure that these settings are saved and then test with various languages to verify that transcription is working as expected. For more detailed guidance, you can refer to the [Multilingual Configuration Guide](https://docs.vapi.ai/customization/multilingual). Additionally, if your setup requires HIPAA compliance, Deepgram is supported in such scenarios alongside Azure, ensuring your transcription meets necessary compliance standards.