open-webui/docker-compose.audio.yaml

services:
  asr:
    build:
      context: ../asr-server
    container_name: asr
    restart: unless-stopped
    ports:
      - '8020:8020'
    environment:
      - 'ASR_AUTH_TOKEN=${ASR_AUTH_TOKEN}'
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  tts:
    build:
      context: ../tts-server
    container_name: tts
    restart: unless-stopped
    ports:
      - '8010:8010'
    environment:
      - 'TTS_AUTH_TOKEN=${TTS_AUTH_TOKEN}'
      - 'HF_HOME=/app/data'
      # Fallback voice for direct /v1/audio/speech calls that use
      # AUDIO_TTS_MODEL=sopro-v2-turbo without a `voice` field. Set to the
      # basename (no .wav) of a clip dropped into services/tts-server/voices/.
      # - 'TTS_SOPRO_DEFAULT_VOICE=your_voice_name'
    volumes:
      # Caches the downloaded Kokoro + Sopro weights across rebuilds.
      - tts-data:/app/data
      # Sopro voice reference clips (<name>.wav); drop files in here on the
      # host and reference them by name via AUDIO_TTS_VOICE / `voice`.
      - ../tts-server/voices:/app/voices:ro
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  open-webui:
    depends_on:
      - asr
      - tts
    environment:
      - 'AUDIO_STT_ENGINE=openai'
      - 'AUDIO_STT_OPENAI_API_BASE_URL=http://asr:8020/v1'
      - 'AUDIO_STT_OPENAI_API_KEY=${ASR_AUTH_TOKEN}'
      - 'AUDIO_STT_MODEL=nemotron-3.5-asr-streaming-0.6b'
      - 'AUDIO_TTS_ENGINE=openai'
      - 'AUDIO_TTS_OPENAI_API_BASE_URL=http://tts:8010/v1'
      - 'AUDIO_TTS_OPENAI_API_KEY=${TTS_AUTH_TOKEN}'
      # To switch back to Kokoro (fixed presets), set:
      #   AUDIO_TTS_MODEL=kokoro-82m
      #   AUDIO_TTS_VOICE=af_heart
      - 'AUDIO_TTS_MODEL=sopro-v2-turbo'
      - 'AUDIO_TTS_VOICE=jarvis-target'

  wyoming-openai:
    image: ghcr.io/roryeckel/wyoming_openai:latest
    container_name: wyoming-openai
    restart: unless-stopped
    depends_on:
      - asr
      - tts
    ports:
      - '10300:10300'
    environment:
      - 'STT_OPENAI_URL=http://asr:8020/v1'
      - 'STT_OPENAI_KEY=${ASR_AUTH_TOKEN}'
      - 'STT_MODELS=nemotron-3.5-asr-streaming-0.6b'
      - 'TTS_OPENAI_URL=http://tts:8010/v1'
      - 'TTS_OPENAI_KEY=${TTS_AUTH_TOKEN}'
      - 'TTS_MODELS=sopro-v2-turbo'
      - 'TTS_VOICES=jarvis-target'
      - 'TTS_SPEED=1.2'

volumes:
  tts-data: {}