ConnectionResetError: [Errno 104] Connection reset by peer

[WesleyFister]

I'm using Wyoming Satellite and when I speak commands I sometimes get this error. It doesn't always happen and in v0.3.6 the commands still get executed (I believe in v0.3.5 the commands were getting dropped).

2025-08-12 21:20:52.216182+00:00INFO:wyoming_openai.handler:Recording started at 16000 Hz, 1 channels, 2 bytes per sample
2025-08-12 21:20:55.291704+00:00INFO:httpx:HTTP Request: POST http://speaches:8000/v1/audio/transcriptions "HTTP/1.1 200 OK"
2025-08-12 21:20:55.292242+00:00INFO:wyoming_openai.handler:Successfully transcribed: Can you turn off the lamp in the room?
2025-08-12 21:20:55.292902+00:00ERROR:asyncio:Task exception was never retrieved
2025-08-12 21:20:55.292921+00:00future: <Task finished name='wyoming event handler' coro=<AsyncEventHandler.run() done, defined at /usr/local/lib/python3.12/site-packages/wyoming/server.py:31> exception=ConnectionResetError(104, 'Connection reset by peer')>
2025-08-12 21:20:55.292946+00:00Traceback (most recent call last):
2025-08-12 21:20:55.292954+00:00File "/usr/local/lib/python3.12/site-packages/wyoming/server.py", line 37, in run
2025-08-12 21:20:55.292961+00:00event = await async_read_event(self.reader)
2025-08-12 21:20:55.292968+00:00^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-08-12 21:20:55.292988+00:00File "/usr/local/lib/python3.12/site-packages/wyoming/event.py", line 84, in async_read_event
2025-08-12 21:20:55.292996+00:00json_line = await reader.readline()
2025-08-12 21:20:55.293003+00:00^^^^^^^^^^^^^^^^^^^^^^^
2025-08-12 21:20:55.293011+00:00File "/usr/local/lib/python3.12/asyncio/streams.py", line 568, in readline
2025-08-12 21:20:55.293023+00:00line = await self.readuntil(sep)
2025-08-12 21:20:55.293031+00:00^^^^^^^^^^^^^^^^^^^^^^^^^
2025-08-12 21:20:55.293040+00:00File "/usr/local/lib/python3.12/asyncio/streams.py", line 660, in readuntil
2025-08-12 21:20:55.293049+00:00await self._wait_for_data('readuntil')
2025-08-12 21:20:55.293066+00:00File "/usr/local/lib/python3.12/asyncio/streams.py", line 545, in _wait_for_data
2025-08-12 21:20:55.293075+00:00await self._waiter
2025-08-12 21:20:55.293082+00:00File "/usr/local/lib/python3.12/asyncio/selector_events.py", line 1013, in _read_ready__data_received
2025-08-12 21:20:55.293102+00:00data = self._sock.recv(self.max_size)
2025-08-12 21:20:55.293110+00:00^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2025-08-12 21:20:55.293117+00:00ConnectionResetError: [Errno 104] Connection reset by peer

This error only seems to happen when you give the Wyoming integration in Home Assistant the URL of the Wyoming OpenAI container like this.

I have not encountered this error by using the IP of the host.

This is my Wyoming OpenAI compose file.

networks:
  speaches_network:
    driver: bridge
    name: speaches_network
  wyoming_openai_network:
    driver: bridge
    name: wyoming_openai_network
services:
  init-speaches:
    command: |
      /bin/bash -c '
        uv tool run speaches-cli model download speaches-ai/Kokoro-82M-v1.0-ONNX
        uv tool run speaches-cli model download rtlingo/mobiuslabsgmbh-faster-whisper-large-v3-turbo
      '
    container_name: init-speaches
    depends_on:
      speaches:
        condition: service_healthy
    deploy:
      resources:
        limits:
          cpus: '1'
          memory: 4G
        reservations:
          devices:
            - capabilities:
                - gpu
              driver: nvidia
    environment:
      - SPEACHES_BASE_URL=http://speaches:8000
    image: ghcr.io/speaches-ai/speaches:latest-cuda-12.4.1
    networks:
      - speaches_network
    runtime: nvidia
    volumes:
      - huggingface-hub:/home/ubuntu/.cache/huggingface/hub
  speaches:
    container_name: speaches
    deploy:
      resources:
        limits:
          cpus: '2'
          memory: 4G
        reservations:
          devices:
            - capabilities:
                - gpu
              driver: nvidia
    environment:
      - enable_ui=True
      - log_level=info
      - WHISPER__TTL=-1
      - WHISPER__MODEL=rtlingo/mobiuslabsgmbh-faster-whisper-large-v3-turbo
      - WHISPER__compute_type=int8_float32
      - export CHAT_COMPLETION_BASE_URL=http://192.168.1.100:9292/v1
      - export CHAT_COMPLETION_API_KEY=doesnotmatterwhatitis
    healthcheck:
      interval: 30s
      retries: 3
      start_period: 40s
      test:
        - CMD
        - curl
        - '-f'
        - http://localhost:8000/health
      timeout: 10s
    image: ghcr.io/speaches-ai/speaches:latest-cuda-12.4.1
    networks:
      - speaches_network
    ports:
      - '8000:8000'
    restart: unless-stopped
    runtime: nvidia
    volumes:
      - huggingface-hub:/home/ubuntu/.cache/huggingface/hub
  wyoming_openai:
    container_name: wyoming_openai
    depends_on:
      speaches:
        condition: service_healthy
    deploy:
      resources:
        limits:
          cpus: '1'
          memory: 1G
    environment:
      STT_BACKEND: SPEACHES
      STT_MODELS: rtlingo/mobiuslabsgmbh-faster-whisper-large-v3-turbo
      STT_OPENAI_URL: http://speaches:8000/v1
      TTS_BACKEND: SPEACHES
      TTS_MODELS: speaches-ai/Kokoro-82M-v1.0-ONNX
      TTS_OPENAI_URL: http://speaches:8000/v1
      WYOMING_LANGUAGES: en
      WYOMING_LOG_LEVEL: INFO
      WYOMING_URI: tcp://0.0.0.0:10300
    image: ghcr.io/roryeckel/wyoming_openai:latest
    networks:
      - speaches_network
      - wyoming_openai_network
    ports:
      - '10300:10300'
    restart: unless-stopped
volumes:
  huggingface-hub: Null

Home Assistant compose file.

networks:
  wyoming_openai_network:
    external: True
services:
  homeassistant:
    container_name: homeassistant
    depends_on:
      - zigbee2mqtt
    image: ghcr.io/home-assistant/home-assistant:stable
    networks:
      - wyoming_openai_network
    ports:
      - '8123:8123'
    privileged: True
    restart: unless-stopped
    volumes:
      - ./hass-config:/config
      - /etc/localtime:/etc/localtime:ro
      - /run/dbus:/run/dbus:ro

[roryeckel]

I think it has to do with multicast dns adding several seconds to the transcription time, which triggers some timeout in Home Assistant to drop the connection to wyoming_openai. Definitely a network or timeout configuration related thing

[0fflineuser]

Hi,
I had the same problem, I have home-assistant behind a nginx reverse proxy.
I fixed it by changing the proxy_read_timeout in the nginx configuration.
It was proxy_read_timeout 10; and I changed it to proxy_read_timeout 300s;, now it works.

[WesleyFister]

Hi, I had the same problem, I have home-assistant behind a nginx reverse proxy. I fixed it by changing the proxy_read_timeout in the nginx configuration. It was proxy_read_timeout 10; and I changed it to proxy_read_timeout 300s;, now it works.

I am not using a reverse proxy so that can't be the issue for me.