Gemini Realtime API error: 1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration

Agents
, , ,
1

1. Error

Receiving 1007 error from Gemini Live API

2. System

Windows 11

3. Context

  • The only model I am using is → REALTIME_MODEL = “gemini-2.5-flash-native-audio-preview-12-2025”
  • Crash occurrence is complicated because in one session there is no crash despite of tool calls and rag querys and in another session the same error occurs during a query, mid-conversation, or due to a tool call(still unknown).
  • My application allows both TEXT and real time AUDIO input/conversation using Gemini Live Api and LiveKit

4. My pyproject.toml

requires-python = ">=3.10,<3.14"
dependencies = [
    "livekit-agents[google]~=1.5",
    "google-genai>=1.30.0",
    "python-dotenv>=1.0.0",
    "flask>=3.1.3",
    "livekit-api>=1.1.0",
    "flask-cors>=6.0.2",
    "llama-index-vector-stores-qdrant>=0.10.1",
    "qdrant-client>=1.18.0",
    "fastembed>=0.8.0",
]

------------------ OR -----------------------

Package Version

fastembed 0.8.0
flask 3.1.3
flask-cors 6.0.2
google-api-core 2.29.0
google-cloud-speech 2.36.0
google-cloud-texttospeech 2.34.0
google-genai 2.14.0
httpcore 1.0.9
httpx 0.28.1
huggingface-hub 1.24.0
livekit 1.1.13
livekit-agents 1.6.6
livekit-api 1.2.0
livekit-local-inference 0.2.6
livekit-plugins-google 1.6.6
livekit-protocol 1.1.21
websockets 15.0.1

5. Here is my entrypoint/session code

server = AgentServer()

# 1. PREWARM: Loads RAG once per worker process into process userdata
# i.e., one RAG pipeline across all sessions
def prewarm(proc: JobProcess):
    logger.info("Prewarming worker: Light startup (deferring RAG load to first request)...")   

server.setup_fnc = prewarm


@server.rtc_session(agent_name="xyz")
async def entrypoint(ctx: agents.JobContext):
    await ctx.connect()

    # Lazy-load RAG engine per worker instance on demand
    if "query_engine" not in ctx.proc.userdata:
        logger.info("Initializing shared RAG query engine on first request...")
        ctx.proc.userdata["query_engine"] = initialize_query_engine(
            collection_name="xyz-index"
        )

    query_engine = ctx.proc.userdata["query_engine"]
    
    participant = await ctx.wait_for_participant()
    web_identity = participant.identity

    logger.info(f"Frontend participant connected: {web_identity}")

    session = AgentSession(
        llm=google.realtime.RealtimeModel(
            model=REALTIME_MODEL, # Strictly only using "gemini-2.5-flash-native-audio-preview-12-2025"
            voice="Callirrhoe",
            modalities=["AUDIO"], # also tried with this line removed
        ),
    )

    agent = MyClassAgent(room=ctx.room, query_engine=query_engine, web_identity=web_identity, agent_session=session)
    logger.info(f"Model: {REALTIME_MODEL}")
    logger.info("Starting AUDIO native session")
    await session.start(
        room=ctx.room,
        agent=agent,
    )

if __name__ == "__main__":
    agents.cli.run_app(server)

6. Error I see in LiveKit’s session’s logs:

Severity
Logger
__mp_main__
Message
Initializing shared RAG query engine on first request...
Attributes
{
code.file.path: "/app/agent.py"
code.function.name: "entrypoint"
code.line.number: 436
room_id: ""
job_id: ""
logger.name: "__mp_main__"
lk.id: ""
}

Severity
Logger
__mp_main__
Message
Frontend participant connected: user-xyz
Attributes
{
code.file.path: "/app/agent.py"
code.function.name: "entrypoint"
code.line.number: 446
room_id: ""
job_id: ""
logger.name: "__mp_main__"
lk.id: ""
}

Severity
Logger
__mp_main__
Message
Model: gemini-2.5-flash-native-audio-preview-12-2025
Attributes
{
code.file.path: "/app/agent.py"
code.function.name: "entrypoint"
code.line.number: 458
room_id: ""
job_id: ""
logger.name: "__mp_main__"
lk.id: ""
}

Severity
Logger
__mp_main__
Message
Starting AUDIO native session
Attributes
{
code.file.path: "/app/agent.py"
code.function.name: "entrypoint"
code.line.number: 459
room_id: ""
job_id: ""
logger.name: "__mp_main__"
lk.id: ""
}

Severity WARN
Logger
livekit.agents
Message
turn_detection is a TurnDetector, but the LLM is a RealtimeModel with server-side turn detection enabled, ignoring the turn_detection setting
Attributes
{
code.file.path: "/app/.venv/lib/python3.12/site-packages/livekit/agents/voice/agent_activity.py"
code.function.name: "_validate_turn_detection"
code.line.number: 279
room_id: ""
job_id: ""
logger.name: "livekit.agents"
lk.id: ""
}

"Severity
Logger
livekit.plugins.google
Message
error in receive task":1007 None. The audio content type ("CONTENT_TYPE_AUDIO) is not supported for this model configuration."{
   "code.file.path":"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py",
   "code.function.name":"_recv_task",
   "code.line.number":1134,
   "exception.type":"APIError",
   "exception.message":"1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.",
   "exception.stacktrace":"Traceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 527, in _receive\n    raw_response = await self._ws.recv(decode=False)\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/websockets/asyncio/connection.py\", line 322, in recv\n    raise self.protocol.close_exc from self.recv_exc\nwebsockets.exceptions.ConnectionClosedError: received 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.; then sent 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n\nDuring handling of the above exception, another exception occurred:\n\nTraceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 1064, in _recv_task\n    async for response in session.receive():\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 446, in receive\n    while result := await self._receive():\n                    ^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 537, in _receive\n    errors.APIError.raise_error(code, reason, None)\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/errors.py\", line 188, in raise_error\n    raise cls(status_code, response_json, response)\ngoogle.genai.errors.APIError: 1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n",
   "room_id":"",
   "job_id":"",
   "logger.name":"livekit.plugins.google",
   "lk.id":""
}

"Severity
Logger
livekit.plugins.google
Message
Gemini Realtime API error":1007 None. The audio content type ("CONTENT_TYPE_AUDIO) is not supported for this model configuration."{
   "code.file.path":"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py",
   "code.function.name":"_main_task",
   "code.line.number":954,
   "exception.type":"APIError",
   "exception.message":"1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.",
   "exception.stacktrace":"Traceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 527, in _receive\n    raw_response = await self._ws.recv(decode=False)\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/websockets/asyncio/connection.py\", line 322, in recv\n    raise self.protocol.close_exc from self.recv_exc\nwebsockets.exceptions.ConnectionClosedError: received 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.; then sent 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n\nDuring handling of the above exception, another exception occurred:\n\nTraceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 944, in _main_task\n    raise err\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 1064, in _recv_task\n    async for response in session.receive():\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 446, in receive\n    while result := await self._receive():\n                    ^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 537, in _receive\n    errors.APIError.raise_error(code, reason, None)\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/errors.py\", line 188, in raise_error\n    raise cls(status_code, response_json, response)\ngoogle.genai.errors.APIError: 1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n",
   "room_id":"",
   "job_id":"",
   "logger.name":"livekit.plugins.google",
   "lk.id":""
}

"Severity
Logger
livekit.plugins.google
Message
Gemini Live closed the session":context exhausted (1007). Reconnecting would replay the same context and fail again; terminating the session.{
   "code.file.path":"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py",
   "code.function.name":"_main_task",
   "code.line.number":962,
   "exception.type":"APIError",
   "exception.message":"1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.",
   "exception.stacktrace":"Traceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 527, in _receive\n    raw_response = await self._ws.recv(decode=False)\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/websockets/asyncio/connection.py\", line 322, in recv\n    raise self.protocol.close_exc from self.recv_exc\nwebsockets.exceptions.ConnectionClosedError: received 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.; then sent 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n\nDuring handling of the above exception, another exception occurred:\n\nTraceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 944, in _main_task\n    raise err\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 1064, in _recv_task\n    async for response in session.receive():\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 446, in receive\n    while result := await self._receive():\n                    ^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 537, in _receive\n    errors.APIError.raise_error(code, reason, None)\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/errors.py\", line 188, in raise_error\n    raise cls(status_code, response_json, response)\ngoogle.genai.errors.APIError: 1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n",
   "room_id":"",
   "job_id":"",
   "logger.name":"livekit.plugins.google",
   "lk.id":""
}

"Severity
Logger
livekit.agents
Message
AgentSession is closing due to unrecoverable error"{
   "code.file.path":"/app/.venv/lib/python3.12/site-packages/livekit/agents/voice/agent_session.py",
   "code.function.name":"_on_error",
   "code.line.number":1636,
   "exception.type":"APIError",
   "exception.message":"1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.",
   "exception.stacktrace":"Traceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 527, in _receive\n    raw_response = await self._ws.recv(decode=False)\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/websockets/asyncio/connection.py\", line 322, in recv\n    raise self.protocol.close_exc from self.recv_exc\nwebsockets.exceptions.ConnectionClosedError: received 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.; then sent 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n\nDuring handling of the above exception, another exception occurred:\n\nTraceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 944, in _main_task\n    raise err\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 1064, in _recv_task\n    async for response in session.receive():\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 446, in receive\n    while result := await self._receive():\n                    ^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 537, in _receive\n    errors.APIError.raise_error(code, reason, None)\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/errors.py\", line 188, in raise_error\n    raise cls(status_code, response_json, response)\ngoogle.genai.errors.APIError: 1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n",
   "room_id":"",
   "job_id":"",
   "logger.name":"livekit.agents",
   "lk.id":""
}

"Severity
Logger
livekit.plugins.google
Message
Error in _main_task"{
   "code.file.path":"/app/.venv/lib/python3.12/site-packages/livekit/agents/utils/log.py",
   "code.function.name":"async_fn_logs",
   "code.line.number":22,
   "exception.type":"APIConnectionError",
   "exception.message":"Gemini Live session context exhausted (1007)",
   "exception.stacktrace":"Traceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 527, in _receive\n    raw_response = await self._ws.recv(decode=False)\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/websockets/asyncio/connection.py\", line 322, in recv\n    raise self.protocol.close_exc from self.recv_exc\nwebsockets.exceptions.ConnectionClosedError: received 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.; then sent 1007 (invalid frame payload data) The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n\nDuring handling of the above exception, another exception occurred:\n\nTraceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 944, in _main_task\n    raise err\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 1064, in _recv_task\n    async for response in session.receive():\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 446, in receive\n    while result := await self._receive():\n                    ^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/live.py\", line 537, in _receive\n    errors.APIError.raise_error(code, reason, None)\n  File \"/app/.venv/lib/python3.12/site-packages/google/genai/errors.py\", line 188, in raise_error\n    raise cls(status_code, response_json, response)\ngoogle.genai.errors.APIError: 1007 None. The audio content type (CONTENT_TYPE_AUDIO) is not supported for this model configuration.\n\nThe above exception was the direct cause of the following exception:\n\nTraceback (most recent call last):\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/agents/utils/log.py\", line 17, in async_fn_logs\n    return await fn(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/app/.venv/lib/python3.12/site-packages/livekit/plugins/google/realtime/realtime_api.py\", line 969, in _main_task\n    raise APIConnectionError(\nlivekit.agents._exceptions.APIConnectionError: Gemini Live session context exhausted (1007)\n",
   "room_id":"",
   "job_id":"",
   "logger.name":"livekit.plugins.google",
   "lk.id":""
}

"Severity
Logger
root
Message
ignoring text stream with topic""lk.chat",
"no callback attached"{
   "code.file.path":"/app/.venv/lib/python3.12/site-packages/livekit/rtc/room.py",
   "code.function.name":"_handle_stream_header",
   "code.line.number":1092,
   "room_id":"",
   "job_id":"",
   "logger.name":"root",
   "lk.id":""
}

Thank you in advance

2

For this line:

modalities=[“AUDIO”], # also tried with this line removed

It is provided if you need to use a separate TTS, which is not what I believe you want to do. See also this caveat in the docs: Gemini Live API plugin | LiveKit Documentation

Can you remove that line, and say what error you see? The logs above all imply to me that the line is still present.

Better still, if you can take the agent starter, GitHub - livekit-examples/agent-starter-python: A complete voice AI starter for LiveKit Agents with Python. · GitHub, and just swap out the realtime as model="google/gemini-2.5-flash-lite", I would expect it to work through LiveKit Inference.

3

Hi Darryn, Thank you for the reply.

Yes I have removed the ‘modalities’ line and run it 3-4 times again and now it is working. I’ll follow up again, just in case the error comes back like it did last time, lol!

And while I have you, could you please suggest ways to reduce latency between the user and agent’s conversation. for example this is how I my application feels:

  1. The AI Agent takes a little too long to respond to user query’s despite of having only 10 datapoints with “similarity_top_k=3” and less than 10 function_tool’s
  2. The user speech detection and text appearance in chat box takes a little too long, like the user’s query sometimes takes 6-8 seconds to be sent to the AI Agent and displayed on the chat box. I am using “useSessionMessages from @livekit/components-react”
  3. my token server and frontend is deployed in US-East and my Livekit Cloud’s AI Agent is deployed in US-East and I(the user) is using it from India(well that’s a constraint)

---------------------------------

And the agent starter link provided implements the half-cascade archi which I am not aiming for at the moment as you guessed correctly, however the commented link provided in that code leads back to gemini live api docs i.e., the Speech-to-Speech archi I am using now..

Thanks!

4

There is no single answer to the question of reducing latency, my go to resource is:

6-8 seconds is way too long, so you should try to understand which part of your flow is responsible for most of that. Most of the advice is written for pipeline, but you should also try different realtime models to see if you get better performance.

If all your users are in India, you might also find this a good read: