livekit.…ns.google Gemini Realtime API error: received 1008 (policy violation) models/gemini-2.5-flash-preview-09-2… is not found for API version v1beta, or is not supported for bidiGenerateContent. C; then sent 1008 (policy violation) models/gemini-2.5-flash-preview-09-2… is not found for API version v1beta, or is not supported for bidiGenerateContent. C Traceback (most recent ca

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

can you provide all available model

You're asking for all available Gemini models for the Realtime API. The LiveKit plugin supports these Gemini realtime models: •

gemini-2.0-flash-exp

•

gemini-2.5-flash-native-audio-preview-09-2025

If you need examples for initializing either model with

google.realtime.RealtimeModel

, or want to confirm which one supports a specific feature (speech-to-speech, native audio input, etc.), tell me which feature you need and I’ll provide the exact snippet. Sources: Gemini Live API plugin | LiveKit docs

wich model will return text as response

You can get text responses from either supported realtime model —

gemini-2.0-flash-exp

or

gemini-2.5-flash-native-audio-preview-09-2025

— by configuring the model to use the TEXT modality. If you want, tell me which SDK/language you’re using and I’ll provide the exact initialization snippet to return text only. Sources: Gemini Live API plugin | LiveKit docs

Error event: type='error' error=RealtimeModelError(type='realtime_model_error', timestamp=1765964195.9799895, label='livekit.plugins.google.realtime.realtime_api.RealtimeModel', error=ConnectionClosedError(Close(code=1007, reason='Cannot extract voices from a non-audio request.'), Close(code=1007, reason='Cannot extract voices from a non-audio request.'), True), recoverable=False) source=<livekit.plugins.google.realtime.realtime_api.RealtimeModel object at 0x000001D5AB1C42F0> created_at=1765964195.980391

You're seeing a

1007

close with reason "Cannot extract voices from a non-audio request." — that means the RealtimeModel expected audio in the request (for voice extraction / native-audio flows) but received a non-audio payload. What you can do next: • If you want text-only responses: configure the model to use the TEXT modality so audio processing/voice extraction isn't required. ◦ Set

modalities=[Modality.TEXT]

(or the equivalent in your SDK) when creating the RealtimeModel. • If you intend to send audio (speech-to-speech or native-audio flows): make sure you send a proper audio track / audio frames on the bidi stream or attach a microphone track so the RealtimeModel receives audio input. • If using a native-audio model (e.g.,

gemini-2.5-flash-native-audio-preview-09-2025

), do not send a non-audio request that includes voice configuration expecting extraction. If you’d like, tell me: 1. Which model you’re using. 2. The modalities/options you set when initializing the RealtimeModel. 3. A short code snippet showing how you open the bidi stream / send the request. I’ll provide the exact snippet to fix either the text-only or audio flow.

model="gemini-2.5-flash-native-audio-preview-09-2025",

You're using

gemini-2.5-flash-native-audio-preview-09-2025

. That model expects native audio unless configured otherwise — your error shows the model received a non-audio request while voice extraction was expected. To resolve this, choose one of two options and tell me which you want to implement: 1) Text-only responses (no audio required) • Initialize the RealtimeModel with TEXT modality so the model won't expect audio:

from livekit.plugins import google
from livekit.plugins.google import Modality

llm = google.realtime.RealtimeModel(
  model="gemini-2.5-flash-native-audio-preview-09-2025",
  modalities=[Modality.TEXT]
)

This accepts text input (or transcribed audio) and returns text responses. 2) Audio (speech-to-speech / native audio) — send real audio • Ensure you send an audio track/frames on the bidi stream (microphone or PCM frames) rather than a non-audio payload. • Do not set modalities to TEXT-only; include audio modality or native audio settings so voice extraction runs. Tell me which option you want (text-only or audio). If audio, paste: • The code snippet that opens the bidi stream and sends the initial request, and • How you attach/send audio (mic track, raw frames, or other). I’ll provide the exact snippet to fix your flow.

llm=google.realtime.RealtimeModel( model="gemini-2.5-flash-native-audio-preview-09-2025", api_key=settings.GOOGLE_API_KEY, modalities=[Modality.TEXT], temperature=0.8, ), i have add this but still got the error

Thanks — I see you’re using

gemini-2.5-flash-native-audio-preview-09-2025

with

modalities=[Modality.TEXT]

but still get the 1007 "Cannot extract voices from a non-audio request" error. Let's pinpoint the issue. Please check and apply these items: 1. Import and use the correct

Modality

enum:

from google.genai.types import Modality

1. Ensure you disable audio in room options so the agent won’t initialize audio input:

from livekit.plugins.room_io import RoomInputOptions
room_options = RoomInputOptions(audio_enabled=False)

1. Remove any voice/tts/affective settings (e.g.,

voice

,

enable_affective_dialog

) when using

modalities=[Modality.TEXT]

. 2. Minimal initialization example to test:

from livekit.agents import AgentSession
from livekit.plugins import google
from google.genai.types import Modality
from livekit.plugins.room_io import RoomInputOptions

session = AgentSession(
  llm=google.realtime.RealtimeModel(
    model="gemini-2.5-flash-native-audio-preview-09-2025",
    api_key=settings.GOOGLE_API_KEY,
    modalities=[Modality.TEXT],
    temperature=0.8,
  ),
  room_options=RoomInputOptions(audio_enabled=False),
)

If you’ve already done all the above and still see the error, please paste: • The exact imports and the small code block that creates the RealtimeModel and starts the session (minimal reproducible snippet). • Any room/agent options you set (especially anything related to audio or TTS). I’ll review that snippet and point out the specific line causing the audio expectation. Sources: Gemini Live API plugin | LiveKit docs | How to Set Up Speech-to-Text Input with Text-Only Output in LiveKit Agents | Text and transcriptions | LiveKit docs | OpenAI Realtime API plugin guide | LiveKit docs

session = AgentSession( llm=google.realtime.RealtimeModel( model="gemini-2.0-flash-exp", api_key=settings.GOOGLE_API_KEY, modalities=[Modality.TEXT], temperature=0.8, ), tts=get_tts(assistant), user_away_timeout=25.0, ) this model is working fine