This message was deleted.

Thanks for reaching out, Durgesh. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

can I pass in multiple languages ?

You cannot pass multiple languages simultaneously. The Gemini Realtime

input_audio_transcription

language field accepts a single ISO-639-1 code (for example

en

,

es

,

fr

) to restrict STT to one language. If you need multi-language support, would you like options for: • detecting language first and then reconfiguring transcription per detected language, or • running parallel transcriptions with separate model instances for each language?

input_audio_transcription=genai_types.AudioTranscription( AttributeError: module 'google.genai.types' has no attribute 'AudioTranscription'

Thanks — I see you're hitting

AttributeError: module 'google.genai.types' has no attribute 'AudioTranscription'

when trying to set

input_audio_transcription

. Let's troubleshoot and fix that. Could you tell me: • which LiveKit / agents plugin package and version you have installed (pip show livekit-agents or contents of requirements.txt)? • the exact import lines you're using now? Quick fixes to try now: • Use the plugin's types import instead of

google.genai.types

. Example:

from livekit.plugins.google.types import AudioTranscription
from livekit.plugins import google
from livekit.agents import AgentSession

session = AgentSession(
    llm=google.realtime.RealtimeModel(
        model="gemini-2.5-flash",
        input_audio_transcription=AudioTranscription(language="es")
    )
)

• If that import still fails, upgrade the package:

pip install --upgrade "livekit-agents[google]"

(ensure version is ~1.3+). Note on language restriction: the plugin supports setting a single ISO-639-1 code via

language

(e.g.,

en

,

es

). Passing multiple languages at once is not supported. If you share your package version and current imports, I’ll give the exact code or next steps. Sources: Gemini Live API plugin | LiveKit docs

The above exception was the direct cause of the following exception: Traceback (most recent call last): File "/home/ubuntu/anaconda3/envs/icall-backend/lib/python3.10/site-packages/livekit/plugins/google/stt.py", line 542, in _run task.result() File "/home/ubuntu/anaconda3/envs/icall-backend/lib/python3.10/site-packages/livekit/plugins/google/stt.py", line 442, in process_stream async for resp in stream: File "/home/ubuntu/anaconda3/envs/icall-backend/lib/python3.10/site-packages/google/api_core/grpc_helpers_async.py", line 109, in _wrapped_aiter raise exceptions.from_grpc_error(rpc_error) from rpc_error google.api_core.exceptions.InvalidArgument: 400 Config contains unsupported fields. See

error_details_ext

extension for details. [field_violations { field: "config.features.enable_automatic_punctuation" description: "Recognizer does not support feature: automatic_punctuation" } ]

Thanks — I see the traceback and the 400 error caused by an unsupported STT field (

enable_automatic_punctuation

) being sent to the Google/Gemini realtime path. What happened: Gemini Realtime uses its own built-in transcription and rejects Google STT config fields (like automatic punctuation). That causes the InvalidArgument error in livekit/plugins/google/stt.py when the plugin forwards an unsupported config. What to do next: • Stop passing Google STT config objects into the Gemini realtime model. Instead set the language via the RealtimeModel parameter: ◦ Python: use

language_code="es-ES"

in

google.realtime.RealtimeModel(...)

◦ Node: use

languageCode: "es-ES"

in

new google.realtime.RealtimeModel({...})

• Remove or avoid

input_audio_transcription

/ external STT config that includes fields such as

enable_automatic_punctuation

. • If you need punctuation or other STT features not supported by Gemini Realtime, run an external STT pipeline separately (not via

input_audio_transcription

) and feed its results into your app. To help further, please share: • The AgentSession / RealtimeModel code you're running (exact snippet). • Output of: ◦

pip show livekit-agents

and

pip show livekit-plugins-google

(or your package manager equivalents). I can then point to the exact code change or patch to remove the unsupported fields. Sources: Gemini Live API plugin | LiveKit docs | Realtime model plugins | LiveKit Docs

its giving error in below code llm = google.LLM( model = model_id, vertexai = vertexai, project = "icalls", location = "us-central1", temperature = temperature, # tool_choice="required", max_output_tokens=200, # Limit for faster responses in voice context top_p=0.95, # Nucleus sampling for quality # automatic_function_calling_config={ # "enabled": True, # "max_iterations": 2, # } ), stt = google.STT( languages=["en-US",language], detect_language=True, interim_results=True, use_streaming=True, sample_rate=16000, # model="long", # automatic_punctuation=False, enable_automatic_punctuation=False, punctuate=False, ), tts = google.TTS( language=language, voice_name=language+"-Chirp3-HD-"+voice_id, # gender=gender, volume_gain_db=2.0, use_streaming=True, pitch=0 # location="us-central1", ),

Thanks — I see the failing code and the root cause. You're passing Google STT config (like

enable_automatic_punctuation

, multiple

languages

,

punctuate

) into the Gemini realtime path, which rejects those fields. Gemini Realtime uses its own built-in STT and expects language via the RealtimeModel

language_code

(BCP-47), not the Google STT plugin config. What to change • Remove or stop passing

google.STT(...)

options into the Gemini realtime model when that model is used for transcription. • Set the language on the RealtimeModel with

language_code

. • If you still need a separate STT instance (for turn detection or a different pipeline), run it independently and do not feed its config fields into the Gemini realtime config. Suggested corrected pattern (Python sketch)

# Use RealtimeModel for Gemini realtime audio + language setting
llm = google.realtime.RealtimeModel(
    model=model_id,
    language_code="en-US",   # set desired BCP-47 language
    temperature=temperature,
    max_output_tokens=200,
    top_p=0.95,
    # no input_audio_transcription / no external STT fields here
)

# If you need an external STT process, run it separately (not passed into Gemini)
external_stt = google.STT(
    languages=["en-US", language],
    detect_language=True,
    interim_results=True,
    use_streaming=True,
    sample_rate=16000,
    # DO NOT pass enable_automatic_punctuation or other Gemini-unsupported fields
)

Quick checklist to avoid the InvalidArgument error • Remove

enable_automatic_punctuation

,

punctuate

, and similar fields from any config that ends up sent to the Gemini realtime plugin. • Do not pass an array of languages to Gemini; set a single

language_code

on the

RealtimeModel

. • If you need punctuation or multi-language transcription, perform that outside Gemini (post-process or separate STT service). If you want, share: • The exact AgentSession / RealtimeModel/llm instantiation code you use now, and •

pip show livekit-agents

and

pip show livekit-plugins-google

outputs I’ll give the minimal patch or code edits to fix the error.