Deepgram Nova-3 multilingual STT: mismatch with pipecat Language enum causing warning

[SwAt1563]

pipecat version

0.0.106

Python version

3.13

Operating System

Docker

Use Case Description

We are using Deepgram Nova-3 STT models with multilingual detection enabled.

According to Deepgram documentation:
https://developers.deepgram.com/docs/models-languages-overview

To enable multi-language detection, the language parameter should be set to "multi".

Current Approach

We pass `"multi"` as the language value when initializing the STT service.

However, in pipecat, the `Language` enum does not support `"multi"` as a valid value. This results in a warning during initialization:


WARNING | pipecat.services.stt_service:__init__:135 - Language string 'multi' is not a recognized Language code. It will be passed to the service as-is.


Previously, this worked without warning when using `LiveOptions`, which supported `"multi"` directly.

Errors or Unexpected Behavior

• Warning is raised when using "multi":

  Language string 'multi' is not a recognized Language code
  

Additional Context

To align with pipecat’s Language enum while still supporting Deepgram multilingual detection:

• Allow language=None in STT settings (do not default to "EN" or any specific language)

• Before sending the request to Deepgram:

  • Convert None → "multi"

• This avoids enum conflicts while preserving intended behavior

[markbackman]

Thanks for reporting this! The warning was overly noisy for valid service-specific language values like "multi". We've downgraded it from a warning to a debug-level log in #4137.

Your current approach of passing language="multi" is correct and works as intended — the value gets passed through to Deepgram as-is. The only issue was the unnecessary warning, which will no longer appear at default log levels.