feat: FRIDAY Voice stack on the GPU machine; VoiceCoach history

FRIDAY Voice (speaches GPU STT/TTS + hermes agent + push-to-talk console) on .150 Docker Desktop, fronted by coach.hobbs.farm. Also captures the VoiceCoach compose/agent trees and deploy notes it replaced, which had never been committed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-11 11:19:04 -05:00
parent dddc6734d0
commit 266ea11447
29 changed files with 4494 additions and 0 deletions
@@ -0,0 +1,30 @@
 # Hermes Agent on .150 (Docker Desktop) — FRIDAY's reasoning agent.
 # Data/settings live in the `hermes-data` named volume, seeded from the user's
 # WSL ~/.hermes (settings preserved; SQLite state.db kept on the VM ext4, not a
 # bridged bind mount). friday-voice calls this via `docker exec hermes hermes ...`.
 services:
  gateway:
    image: nousresearch/hermes-agent:latest
    container_name: hermes
    restart: unless-stopped
    volumes:
      - hermes-data:/opt/data
    environment:
      - HERMES_UID=10000
      - HERMES_GID=10000
    command: ["gateway", "run"]
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
  # dashboard: disabled — crash-loops under s6 on Docker Desktop and not needed
  # by friday-voice (gateway + `docker exec hermes` is all that's required).
  # Re-enable + debug later if the localhost:9119 web UI is wanted.
 volumes:
  hermes-data:
    external: true
@@ -0,0 +1,36 @@
 services:
  speaches:
    image: ghcr.io/speaches-ai/speaches:latest-cuda
    container_name: voicecoach-speaches-gpu
    restart: unless-stopped
    environment:
      - WHISPER__INFERENCE_DEVICE=cuda
      - WHISPER__COMPUTE_TYPE=float16
      - WHISPER__TTL=-1
    volumes:
      - speaches-models:/home/ubuntu/.cache
    ports:
      - "8000:8000"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 180s
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
 volumes:
  speaches-models:
@@ -0,0 +1,9 @@
 # VoiceCoach Environment Variables
 # Copy to .env and fill in values
 # Anthropic (required - for Claude LLM)
 ANTHROPIC_API_KEY=sk-ant-...
 # LiveKit (auto-generated on first deploy if not set)
 LIVEKIT_API_KEY=voicecoach-key
 LIVEKIT_API_SECRET=your-livekit-secret-here
@@ -0,0 +1,184 @@
 services:
  # LiveKit WebRTC Server (unchanged)
  livekit-server:
    image: livekit/livekit-server:latest
    container_name: voicecoach-livekit
    restart: unless-stopped
    command:
      - --config
      - /etc/livekit/livekit.yaml
      - --keys
      - "${LIVEKIT_API_KEY}: ${LIVEKIT_API_SECRET}"
    ports:
      - "7880:7880"
      - "7881:7881"
      - "50000-50100:50000-50100/udp"
    volumes:
      - ./livekit/livekit.yaml:/etc/livekit/livekit.yaml:ro
    dns_search: [""]
    networks:
      - voicecoach
    mem_limit: 512M
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
  # Speaches: self-hosted Whisper STT + Kokoro TTS (v2 - replaces Piper + in-process Whisper)
  # NOTE: No WHISPER__MODEL or TTS__* env vars - WhisperConfig has no 'model' field in this version.
  # Models are loaded on-demand per-request (model ID passed in API body) or pre-downloaded via
  # POST /v1/models/{model_id}. TTS model/voice are also per-request in POST /v1/audio/speech body.
  # Volume must map to /home/ubuntu/.cache (container runs as 'ubuntu', not root).
  speaches:
    image: ghcr.io/speaches-ai/speaches:latest-cpu
    container_name: voicecoach-speaches
    restart: unless-stopped
    environment:
      - WHISPER__INFERENCE_DEVICE=cpu
      - WHISPER__COMPUTE_TYPE=int8
      - WHISPER__TTL=-1
    volumes:
      - /mnt/docker-data/configs/voicecoach/speaches-models:/home/ubuntu/.cache
    dns_search: [""]
    networks:
      - voicecoach
    mem_limit: 3G
    memswap_limit: 3G
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:8000/health"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 120s
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
  # SpeechBrain Emotion Analysis (unchanged, internal only)
  speechbrain-service:
    build:
      context: ./speechbrain-service
    image: voicecoach-speechbrain:latest
    container_name: voicecoach-speechbrain
    restart: unless-stopped
    volumes:
      - /mnt/docker-data/configs/voicecoach/speechbrain-models:/app/models
      - /mnt/docker-data/configs/voicecoach/clips:/data/clips
    environment:
      - MODEL_DIR=/app/models
    dns_search: [""]
    networks:
      - voicecoach
    mem_limit: 3G
    memswap_limit: 3G
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
  # OpenClaw VoiceCoach Agent (v2 - LLM gateway with memory/skills)
  voicecoach-openclaw:
    image: margaret-openclaw:latest
    container_name: voicecoach-openclaw
    restart: unless-stopped
    user: "0:0"
    ports:
      - "3100:3100"
    environment:
      - NODE_ENV=production
      - HOME=/home/glyptic
      - USER=glyptic
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
      - OPENCLAW_STATE_DIR=/home/glyptic/.voicecoach
      - OPENCLAW_CONFIG_PATH=/home/glyptic/.voicecoach/openclaw.json
    volumes:
      - /mnt/infrastructure/voicecoach:/home/glyptic/.voicecoach
    command: ["sh", "-c", "mkdir -p /home/glyptic && node --max-old-space-size=1500 /usr/local/lib/node_modules/openclaw/dist/index.js gateway --allow-unconfigured"]
    dns_search: [""]
    networks:
      - voicecoach
    mem_limit: 2G
    memswap_limit: 2G
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
  # Pipecat Voice Pipeline Agent (v2 - uses Speaches + OpenClaw gateway)
  pipecat-agent:
    build:
      context: ./pipecat-agent
    image: voicecoach-pipecat:latest
    container_name: voicecoach-pipecat
    restart: unless-stopped
    env_file:
      - .env
    environment:
      - LIVEKIT_URL=ws://livekit-server:7880
      - LIVEKIT_PUBLIC_URL=wss://voice.hobbs.farm
      - SPEACHES_URL=http://192.168.86.150:8000
      - SPEECHBRAIN_URL=http://speechbrain-service:8001
      - OPENCLAW_URL=http://voicecoach-openclaw:3100
      - OPENCLAW_AGENT_ID=voicecoach
      - WHISPER_MODEL=deepdml/faster-whisper-large-v3-turbo-ct2
      - KOKORO_VOICE=af_nova
      - CLIPS_DIR=/data/clips/pending
    volumes:
      - /mnt/docker-data/configs/voicecoach/clips:/data/clips
    depends_on:
      - livekit-server
      - speechbrain-service
      - voicecoach-openclaw
    dns_search: [""]
    networks:
      - voicecoach
    mem_limit: 1G
    memswap_limit: 1G
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    logging:
      driver: "json-file"
      options:
        max-size: "50m"
        max-file: "3"
  # Web Client (unchanged)
  voicecoach-web:
    build:
      context: ./web-client
    image: voicecoach-web:latest
    container_name: voicecoach-web
    restart: unless-stopped
    ports:
      - "8085:80"
    depends_on:
      - pipecat-agent
    dns_search: [""]
    networks:
      - voicecoach
    mem_limit: 128M
    labels:
      - "com.centurylinklabs.watchtower.enable=false"
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
 networks:
  voicecoach:
    driver: bridge
@@ -0,0 +1,15 @@
 port: 7880
 rtc:
  tcp_port: 7881
  port_range_start: 50000
  port_range_end: 60000
  use_external_ip: true
 logging:
  json: false
  level: info
 room:
  max_participants: 5
  empty_timeout: 300
  departure_timeout: 30
@@ -0,0 +1,63 @@
 You are VoiceCoach, a personal communication and leadership coach for Jeremy.
 You engage with Jeremy during his daily commute via real-time voice conversation.
 ## Conversation Style
 - Keep responses to 2-4 spoken sentences unless Jeremy asks for more
 - Use natural, conversational language -- no bullet points, no markdown
 - You are a coach, not a chatbot. Ask questions. Challenge thinking. Give direct feedback.
 ## Vocal State Analysis
 Before each user message, you will receive a [VOCAL_STATE] tag:
 ```
 [VOCAL_STATE emotion=frustrated confidence=0.82 pitch=215hz energy=0.71 rate=4.8ops voiced=0.91]
 ```
 Use this to coach on HOW Jeremy communicates, not just WHAT he says:
 - High pitch + high energy + fast rate = stressed, rushed -- note this
 - Low voiced fraction = hesitant, uncertain -- worth exploring
 - Elevated pitch from baseline = emotional arousal -- name it gently
 - Never quote the raw numbers. Translate to natural observations.
 - Example: "I can hear some tension in how you're describing that" not "your pitch was 215hz"
 Only comment on vocal state when it's meaningfully different from baseline or
 when it's directly relevant to the coaching topic. Don't narrate every turn.
 ## Coaching Modes
 Jeremy can switch modes mid-conversation. Honor the switch immediately.
 **coaching mode** (default): Active feedback on both content and delivery.
  Ask probing questions. Notice patterns. Give direct recommendations.
 **vent mode**: Listen and reflect. Minimal advice. Validate feelings.
  Mirror back what you hear. Let Jeremy process out loud.
 **practice mode**: Roleplay a difficult stakeholder, colleague, or situation.
  Stay in character. Be realistic -- not easy. Give feedback after.
 **briefing mode**: Summarize Jeremy's current tasks, context, and priorities
  from memory. Be crisp. Flag anything time-sensitive.
 ## Memory Usage
 You have access to:
 - `vocal_baseline`: Jeremy's neutral voice characteristics for comparison
 - `recent_sessions`: Last 3 session summaries with topics and vocal trends
 - `action_items`: Commitments Jeremy made in previous sessions
 - `coaching_themes`: Recurring themes across sessions (track these actively)
 - `jeremy_context`: Jeremy's role, team, current challenges
 At the end of each session, write a summary to `recent_sessions` including:
 - Topics discussed
 - Vocal patterns observed
 - Action items for next session
 - Any new coaching themes identified
 ## Baseline Calibration (First Session)
 If `vocal_baseline` is empty, the first 60 seconds should be calibration.
 Say: "Before we dive in, let's do a quick baseline. Can you tell me about
 your commute in to work today -- just talk naturally for about a minute."
 Then write the averaged vocal readings to `vocal_baseline`.
 ## Things Jeremy Is Working On
 Read from `coaching_themes` and `jeremy_context` in memory.
 Reference previous sessions naturally: "Last week you mentioned..."
 Track whether previous action items were attempted.
@@ -0,0 +1,31 @@
 FROM python:3.11-slim
 RUN apt-get update && apt-get install -y libsndfile1 ffmpeg && \
    rm -rf /var/lib/apt/lists/*
 WORKDIR /app
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 # Patch pipecat LiveKit transport bugs:
 # 1. Remove duplicate on_connected call in connect() that causes InvalidStateError
 # 2. Remove redundant push_frame that bypasses VAD
 RUN python3 -c "\
 import pipecat.transports.services.livekit as m; \
 path = m.__file__; \
 content = open(path).read(); \
 # Fix 1: Wrap the direct on_connected call to prevent InvalidStateError \
 content = content.replace( \
    'await self._callbacks.on_connected()', \
    'try:\n                    await self._callbacks.on_connected()\n                except Exception:\n                    pass  # Already handled by room event', \
    1 \
 ); \
 open(path, 'w').write(content); \
 print(f'Patched {path}')"
 COPY . .
 HEALTHCHECK --interval=30s --timeout=10s --start-period=30s \
    CMD python -c "import httpx; httpx.get('http://localhost:8080/health').raise_for_status()"
 CMD ["python", "main.py"]
@@ -0,0 +1,113 @@
 """
 VoiceCoach pipeline entry point -- v2.
 Runs token server + restarts pipeline immediately after each session.
 """
 import asyncio
 from datetime import timedelta
 import logging
 import os
 import signal
 import sys
 import threading
 import uuid
 import structlog
 import uvicorn
 from fastapi import FastAPI
 from pydantic import BaseModel
 from livekit import api as livekit_api
 from pipeline import run_voice_coach_pipeline
 logging.basicConfig(level=logging.INFO)
 logger = structlog.get_logger()
 LIVEKIT_API_KEY = os.environ["LIVEKIT_API_KEY"]
 LIVEKIT_API_SECRET = os.environ["LIVEKIT_API_SECRET"]
 LIVEKIT_WS_URL = os.getenv("LIVEKIT_URL", "ws://livekit-server:7880")
 LIVEKIT_PUBLIC_URL = os.getenv("LIVEKIT_PUBLIC_URL", "wss://voice.hobbs.farm")
 OPENCLAW_TOKEN = os.getenv("OPENCLAW_TOKEN", "")
 # --- Token HTTP server ---
 token_app = FastAPI(title="VoiceCoach Token Service")
 class TokenRequest(BaseModel):
    identity: str = "jeremy"
 class TokenResponse(BaseModel):
    token: str
    url: str
@token_app.get("/health")
 async def health():
    return {"status": "ok", "version": "2.0"}
@token_app.post("/token")
 async def get_token(req: TokenRequest):
    token = livekit_api.AccessToken(LIVEKIT_API_KEY, LIVEKIT_API_SECRET)
    token.with_grants(livekit_api.VideoGrants(
        room_join=True, room="voicecoach",
        can_publish=True, can_subscribe=True,
    ))
    token.identity = req.identity
    return TokenResponse(token=token.to_jwt(), url=LIVEKIT_PUBLIC_URL)
 def run_token_server():
    uvicorn.run(token_app, host="0.0.0.0", port=8080, log_level="info")
 # --- Main: restart entire process after each session ---
 # This guarantees a completely fresh pipeline for every session.
 # No stale transport state, no cancelled tasks, no dead queues.
 def run_pipeline_sync():
    """Run one pipeline session, then exit the process.
    Docker's restart: unless-stopped will bring us back immediately.
    """
    agent_identity = f"voicecoach-agent-{uuid.uuid4().hex[:8]}"
    token = livekit_api.AccessToken(LIVEKIT_API_KEY, LIVEKIT_API_SECRET)
    token.with_grants(livekit_api.VideoGrants(room_join=True, room="voicecoach"))
    token.identity = agent_identity
    token.ttl = timedelta(hours=24)  # 24 hours - prevents expired token reconnect failures
    livekit_token = token.to_jwt()
    logger.info("Starting VoiceCoach v2 pipeline", identity=agent_identity)
    loop = asyncio.new_event_loop()
    asyncio.set_event_loop(loop)
    try:
        loop.run_until_complete(run_voice_coach_pipeline(
            livekit_url=LIVEKIT_WS_URL,
            livekit_token=livekit_token,
            openclaw_token=OPENCLAW_TOKEN,
        ))
    except Exception as e:
        logger.error("Pipeline ended", error=str(e))
    finally:
        loop.close()
 if __name__ == "__main__":
    # Start token server in background
    server_thread = threading.Thread(target=run_token_server, daemon=True)
    server_thread.start()
    logger.info("Token server started on :8080")
    # Run pipeline sessions in a loop
    # Each iteration creates a completely fresh asyncio event loop and pipeline
    while True:
        try:
            run_pipeline_sync()
        except Exception as e:
            logger.error("Pipeline crash", error=str(e))
        # Brief pause to let LiveKit clean up the old participant
        import time
        time.sleep(3)
        logger.info("Restarting pipeline with fresh state...")
@@ -0,0 +1,34 @@
 """
 Piper TTS client for Pipecat pipeline.
 Calls the self-hosted Piper HTTP service.
 """
 import io
 import logging
 import wave
 import httpx
 logger = logging.getLogger(__name__)
 class PiperTTSClient:
    """HTTP client for the Piper TTS service."""
    def __init__(self, base_url: str = "http://piper-tts:8003"):
        self.base_url = base_url.rstrip("/")
        self._client = httpx.AsyncClient(timeout=10.0)
    async def synthesize(self, text: str) -> bytes:
        """Convert text to WAV audio bytes via Piper service."""
        try:
            response = await self._client.post(
                f"{self.base_url}/synthesize",
                json={"text": text, "length_scale": 0.95, "sentence_silence": 0.15},
            )
            response.raise_for_status()
            return response.content
        except Exception as e:
            logger.error(f"Piper TTS failed: {e}")
            return b""
    async def close(self):
        await self._client.aclose()
@@ -0,0 +1,76 @@
 """
 Builds Claude prompts with embedded vocal state metadata.
 """
 from dataclasses import dataclass
 from typing import Optional
 COACHING_SYSTEM_PROMPT = """You are a personal voice coach and communication advisor named VoiceCoach.
 You are speaking to Jeremy during his commute via voice -- keep responses conversational and concise (2-4 sentences unless asked for more).
 You receive a [VOCAL_STATE] tag before each user message containing paralinguistic analysis of Jeremy's voice:
 - emotion: the detected emotional state
 - confidence: how confident the analysis is (0-1)
 - pitch: average fundamental frequency in Hz (higher = more animated/stressed)
 - energy: vocal energy level (0-1, higher = louder/more forceful)
 - rate: speaking rate in onset events/sec (higher = faster speech)
 - voiced: fraction of voiced frames (lower = more hesitant/uncertain)
 Use this information to provide coaching on BOTH what Jeremy says AND how he sounds saying it.
 Important: Never quote the raw numbers. Translate them into natural observations.
 Example: "I notice some tension in your voice..." not "Your pitch was 215hz..."
 {baseline_context}
 {session_history}
 Available coaching modes (Jeremy can switch by saying the phrase):
 - "coaching mode" (default) -- active feedback on content and delivery
 - "vent mode" -- listen and reflect, minimal advice
 - "practice mode" -- roleplay a difficult stakeholder
 - "briefing mode" -- summarize Jeremy's tasks and context from memory
 Current mode: {current_mode}
 """
@dataclass
 class VocalState:
    emotion: str
    confidence: float
    pitch_mean_hz: Optional[float]
    energy_rms: float
    speaking_rate_approx: float
    voiced_fraction: float
 def format_vocal_state(state: VocalState) -> str:
    pitch_str = f"{state.pitch_mean_hz:.0f}hz" if state.pitch_mean_hz else "unknown"
    return (
        f"[VOCAL_STATE "
        f"emotion={state.emotion} "
        f"confidence={state.confidence:.2f} "
        f"pitch={pitch_str} "
        f"energy={state.energy_rms:.2f} "
        f"rate={state.speaking_rate_approx:.1f}ops "
        f"voiced={state.voiced_fraction:.2f}]"
    )
 def build_user_message(transcript: str, vocal_state: Optional[VocalState]) -> str:
    if vocal_state and vocal_state.confidence > 0.3:
        tag = format_vocal_state(vocal_state)
        return f"{tag}\n{transcript}"
    return transcript
 def build_system_prompt(
    baseline_context: str = "",
    session_history: str = "",
    current_mode: str = "coaching mode",
 ) -> str:
    return COACHING_SYSTEM_PROMPT.format(
        baseline_context=baseline_context or "No baseline established yet. First session: ask Jeremy to read a short neutral calibration script.",
        session_history=session_history or "No previous session history.",
        current_mode=current_mode,
    )
@@ -0,0 +1,7 @@
 pipecat-ai[livekit,silero,openai]==0.0.48
 httpx==0.27.0
 python-dotenv==1.0.1
 structlog==24.4.0
 livekit-api==0.7.1
 fastapi==0.115.0
 uvicorn[standard]==0.30.0
@@ -0,0 +1,35 @@
 """
 Local STT service using faster-whisper.
 Replaces Deepgram with fully self-hosted speech-to-text.
 """
 import io
 import logging
 import wave
 import numpy as np
 from faster_whisper import WhisperModel
 logger = logging.getLogger(__name__)
 class LocalWhisperSTT:
    """Wraps faster-whisper for use in the Pipecat pipeline."""
    def __init__(self, model_size: str = "medium", device: str = "cpu", compute_type: str = "int8"):
        logger.info(f"Loading faster-whisper model: {model_size} on {device}")
        self.model = WhisperModel(model_size, device=device, compute_type=compute_type)
        logger.info("faster-whisper model loaded.")
    def transcribe(self, audio_bytes: bytes, sample_rate: int = 16000) -> str:
        """Transcribe raw PCM audio bytes to text."""
        audio_array = np.frombuffer(audio_bytes, dtype=np.int16).astype(np.float32) / 32768.0
        segments, info = self.model.transcribe(
            audio_array,
            beam_size=5,
            language="en",
            vad_filter=True,
            vad_parameters=dict(min_silence_duration_ms=500),
        )
        text = " ".join(seg.text.strip() for seg in segments)
        return text
@@ -0,0 +1,27 @@
 FROM python:3.11-slim
 RUN apt-get update && apt-get install -y \
    libsndfile1 \
    wget \
    && rm -rf /var/lib/apt/lists/*
 WORKDIR /app
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 # Download Piper voice model at build time from HuggingFace (lessac medium - good quality, fast on CPU)
 RUN mkdir -p /app/models && \
    wget -q "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium.onnx?download=true" \
        -O /app/models/en_US-lessac-medium.onnx && \
    wget -q "https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/en_US-lessac-medium.onnx.json?download=true" \
        -O /app/models/en_US-lessac-medium.onnx.json
 COPY . .
 EXPOSE 8003
 HEALTHCHECK --interval=30s --timeout=10s --start-period=30s \
    CMD python -c "import httpx; httpx.get('http://localhost:8003/health').raise_for_status()"
 CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8003", "--workers", "1"]
@@ -0,0 +1,72 @@
 """
 Piper TTS HTTP service.
 Self-hosted text-to-speech, internal network only.
 """
 import io
 import logging
 import os
 import wave
 from contextlib import asynccontextmanager
 from fastapi import FastAPI, HTTPException
 from fastapi.responses import Response
 from pydantic import BaseModel
 import piper
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 MODEL_PATH = os.getenv("PIPER_MODEL", "/app/models/en_US-lessac-medium.onnx")
 voice = None
@asynccontextmanager
 async def lifespan(app: FastAPI):
    global voice
    logger.info(f"Loading Piper model: {MODEL_PATH}")
    voice = piper.PiperVoice.load(MODEL_PATH)
    logger.info("Piper TTS ready.")
    yield
    logger.info("Shutting down.")
 app = FastAPI(title="Piper TTS Service", lifespan=lifespan)
 class TTSRequest(BaseModel):
    text: str
    speaker_id: int = None
    length_scale: float = 1.0
    sentence_silence: float = 0.2
@app.get("/health")
 async def health():
    if voice is None:
        raise HTTPException(status_code=503, detail="Model not loaded")
    return {"status": "ok", "model": os.path.basename(MODEL_PATH)}
@app.post("/synthesize")
 async def synthesize(req: TTSRequest):
    if voice is None:
        raise HTTPException(status_code=503, detail="Model not loaded")
    if not req.text.strip():
        raise HTTPException(status_code=400, detail="Empty text")
    buf = io.BytesIO()
    with wave.open(buf, "wb") as wf:
        synth_kwargs = {
            "length_scale": req.length_scale,
            "sentence_silence": req.sentence_silence,
        }
        if req.speaker_id is not None:
            synth_kwargs["speaker_id"] = req.speaker_id
        voice.synthesize(req.text, wf, **synth_kwargs)
    return Response(
        content=buf.getvalue(),
        media_type="audio/wav",
        headers={"Content-Disposition": "inline; filename=speech.wav"},
    )
@@ -0,0 +1,5 @@
 fastapi==0.115.0
 uvicorn[standard]==0.30.0
 piper-tts==1.2.0
 httpx==0.27.0
 numpy==1.26.4
@@ -0,0 +1,33 @@
 FROM python:3.11-slim
 RUN apt-get update && apt-get install -y \
    libsndfile1 \
    ffmpeg \
    git \
    && rm -rf /var/lib/apt/lists/*
 WORKDIR /app
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 # Pre-download the SpeechBrain model at build time
 RUN python -c "\
 from speechbrain.inference.interfaces import foreign_class; \
 foreign_class( \
    source='speechbrain/emotion-recognition-wav2vec2-IEMOCAP', \
    pymodule_file='custom_interface.py', \
    classname='CustomEncoderWav2vec2Classifier', \
    savedir='/app/models', \
    run_opts={'device': 'cpu'}, \
 ); \
 print('Model downloaded successfully.')"
 COPY . .
 EXPOSE 8001
 HEALTHCHECK --interval=30s --timeout=10s --start-period=90s \
    CMD python -c "import httpx; httpx.get('http://localhost:8001/health').raise_for_status()"
 CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8001", "--workers", "1"]
@@ -0,0 +1,119 @@
 """
 SpeechBrain + librosa vocal analysis.
 Performs emotion classification and acoustic feature extraction from audio.
 """
 import io
 import logging
 import numpy as np
 import soundfile as sf
 import librosa
 from speechbrain.inference.interfaces import foreign_class
 from dataclasses import dataclass, asdict
 from typing import Optional
 logger = logging.getLogger(__name__)
 SAMPLE_RATE = 16000
 MIN_DURATION_SECONDS = 0.5
@dataclass
 class VocalAnalysis:
    emotion: str
    emotion_confidence: float
    emotion_scores: dict
    pitch_mean_hz: Optional[float]
    pitch_variance: Optional[float]
    energy_rms: float
    speaking_rate_approx: float
    voiced_fraction: float
    duration_seconds: float
    analysis_error: Optional[str] = None
 class VocalAnalyzer:
    """Loads models once at startup and provides analysis per audio chunk."""
    def __init__(self, model_dir: str = "/app/models"):
        logger.info("Loading SpeechBrain emotion model...")
        self.classifier = foreign_class(
            source="speechbrain/emotion-recognition-wav2vec2-IEMOCAP",
            pymodule_file="custom_interface.py",
            classname="CustomEncoderWav2vec2Classifier",
            savedir=model_dir,
            run_opts={"device": "cpu"},
        )
        logger.info("SpeechBrain model loaded successfully.")
    def analyze(self, audio_bytes: bytes, content_type: str = "audio/wav") -> VocalAnalysis:
        try:
            audio_buf = io.BytesIO(audio_bytes)
            audio, sr = sf.read(audio_buf, dtype="float32")
            if audio.ndim > 1:
                audio = audio.mean(axis=1)
            if sr != SAMPLE_RATE:
                audio = librosa.resample(audio, orig_sr=sr, target_sr=SAMPLE_RATE)
                sr = SAMPLE_RATE
            duration = len(audio) / sr
            if duration < MIN_DURATION_SECONDS:
                return VocalAnalysis(
                    emotion="neutral", emotion_confidence=0.0, emotion_scores={},
                    pitch_mean_hz=None, pitch_variance=None, energy_rms=0.0,
                    speaking_rate_approx=0.0, voiced_fraction=0.0,
                    duration_seconds=duration,
                    analysis_error=f"Audio too short: {duration:.2f}s < {MIN_DURATION_SECONDS}s minimum",
                )
            import tempfile, os
            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
                sf.write(tmp.name, audio, sr)
                tmp_path = tmp.name
            try:
                out_prob, score, index, text_lab = self.classifier.classify_file(tmp_path)
                emotion = text_lab[0] if isinstance(text_lab, list) else text_lab
                confidence = float(score[0]) if hasattr(score, "__iter__") else float(score)
                labels = ["neu", "hap", "sad", "ang", "fru"]
                prob_list = out_prob[0].tolist() if hasattr(out_prob[0], "tolist") else list(out_prob[0])
                emotion_scores = {l: round(p, 4) for l, p in zip(labels, prob_list)}
            finally:
                os.unlink(tmp_path)
            f0, voiced_flag, _ = librosa.pyin(
                audio, fmin=librosa.note_to_hz("C2"), fmax=librosa.note_to_hz("C7"), sr=sr,
            )
            voiced_f0 = f0[voiced_flag] if voiced_flag is not None else np.array([])
            pitch_mean = float(np.mean(voiced_f0)) if len(voiced_f0) > 0 else None
            pitch_var = float(np.var(voiced_f0)) if len(voiced_f0) > 1 else None
            voiced_fraction = float(np.sum(voiced_flag) / len(voiced_flag)) if voiced_flag is not None else 0.0
            rms = librosa.feature.rms(y=audio)
            energy_rms = float(np.mean(rms))
            onset_frames = librosa.onset.onset_detect(y=audio, sr=sr)
            speaking_rate = len(onset_frames) / duration if duration > 0 else 0.0
            return VocalAnalysis(
                emotion=emotion, emotion_confidence=round(confidence, 4),
                emotion_scores=emotion_scores,
                pitch_mean_hz=round(pitch_mean, 2) if pitch_mean else None,
                pitch_variance=round(pitch_var, 2) if pitch_var else None,
                energy_rms=round(energy_rms, 4),
                speaking_rate_approx=round(speaking_rate, 2),
                voiced_fraction=round(voiced_fraction, 4),
                duration_seconds=round(duration, 3),
            )
        except Exception as e:
            logger.exception(f"Analysis failed: {e}")
            return VocalAnalysis(
                emotion="neutral", emotion_confidence=0.0, emotion_scores={},
                pitch_mean_hz=None, pitch_variance=None, energy_rms=0.0,
                speaking_rate_approx=0.0, voiced_fraction=0.0,
                duration_seconds=0.0, analysis_error=str(e),
            )
@@ -0,0 +1,66 @@
 """
 FastAPI service wrapping VocalAnalyzer.
 Internal-only -- not exposed via Traefik.
 """
 import logging
 import os
 from contextlib import asynccontextmanager
 from fastapi import FastAPI, UploadFile, File, HTTPException
 from fastapi.responses import JSONResponse
 from dataclasses import asdict
 from analyzer import VocalAnalyzer
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 MODEL_DIR = os.getenv("MODEL_DIR", "/app/models")
 analyzer: VocalAnalyzer = None
@asynccontextmanager
 async def lifespan(app: FastAPI):
    global analyzer
    logger.info("Starting VocalAnalyzer...")
    analyzer = VocalAnalyzer(model_dir=MODEL_DIR)
    logger.info("VocalAnalyzer ready.")
    yield
    logger.info("Shutting down.")
 app = FastAPI(title="SpeechBrain Vocal Analysis Service", lifespan=lifespan)
@app.get("/health")
 async def health():
    if analyzer is None:
        raise HTTPException(status_code=503, detail="Model not loaded")
    return {"status": "ok", "model": "emotion-recognition-wav2vec2-IEMOCAP"}
@app.post("/analyze")
 async def analyze(audio: UploadFile = File(...)):
    if analyzer is None:
        raise HTTPException(status_code=503, detail="Model not loaded")
    content = await audio.read()
    if len(content) == 0:
        raise HTTPException(status_code=400, detail="Empty audio file")
    result = analyzer.analyze(content, content_type=audio.content_type or "audio/wav")
    return JSONResponse(content=asdict(result))
@app.post("/calibrate")
 async def calibrate(audio: UploadFile = File(...)):
    if analyzer is None:
        raise HTTPException(status_code=503, detail="Model not loaded")
    content = await audio.read()
    if len(content) == 0:
        raise HTTPException(status_code=400, detail="Empty audio file")
    result = analyzer.analyze(content, content_type=audio.content_type or "audio/wav")
    result_dict = asdict(result)
    result_dict["is_baseline"] = True
    return JSONResponse(content=result_dict)
@@ -0,0 +1,12 @@
 fastapi==0.115.0
 uvicorn[standard]==0.30.0
 speechbrain==1.0.0
 huggingface_hub==0.19.4
 transformers==4.40.2
 torch==2.3.0
 torchaudio==2.3.0
 librosa==0.10.2
 soundfile==0.12.1
 numpy==1.26.4
 httpx==0.27.0
 python-multipart==0.0.9
@@ -0,0 +1,8 @@
 FROM nginx:alpine
 COPY index.html /usr/share/nginx/html/
 COPY app.js /usr/share/nginx/html/
 COPY style.css /usr/share/nginx/html/
 COPY nginx.conf /etc/nginx/conf.d/default.conf
 EXPOSE 80
@@ -0,0 +1,158 @@
 const LIVEKIT_URL = "wss://voice.hobbs.farm";
 let room = null;
 function log(msg, level) {
  const el = document.getElementById("logOutput");
  const time = new Date().toLocaleTimeString();
  const cls = level === "error" ? "log-error" : level === "warn" ? "log-warn" : "log-info";
  el.innerHTML += `<div class="${cls}"><span class="log-time">${time}</span> ${msg}</div>`;
  el.scrollTop = el.scrollHeight;
 }
 function clearLog() {
  document.getElementById("logOutput").innerHTML = "";
 }
 async function connect() {
  setStatus("connecting", "Connecting...");
  log("Requesting token from server...");
  try {
    const response = await fetch("/api/token", {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ identity: "jeremy-" + Date.now() }),
    });
    if (!response.ok) {
      const text = await response.text();
      throw new Error("Token request failed: " + response.status + " " + text);
    }
    const data = await response.json();
    log("Token received. LiveKit URL: " + data.url);
    room = new LivekitClient.Room({
      audioCaptureDefaults: {
        echoCancellation: true,
        noiseSuppression: true,
        autoGainControl: true,
      },
    });
    room.on(LivekitClient.RoomEvent.Connected, () => {
      setStatus("connected", "Connected");
      document.getElementById("connectBtn").style.display = "none";
      document.getElementById("disconnectBtn").style.display = "block";
      log("Connected to LiveKit room");
      log("Participants in room: " + room.remoteParticipants.size);
      if (room.remoteParticipants.size === 0) {
        log("No agent participant yet - waiting for voicecoach-agent to join...", "warn");
      }
    });
    room.on(LivekitClient.RoomEvent.Disconnected, (reason) => {
      setStatus("disconnected", "Disconnected");
      document.getElementById("connectBtn").style.display = "block";
      document.getElementById("disconnectBtn").style.display = "none";
      log("Disconnected from LiveKit" + (reason ? ": " + reason : ""), "warn");
    });
    room.on(LivekitClient.RoomEvent.ParticipantConnected, (participant) => {
      log("Participant joined: " + participant.identity);
    });
    room.on(LivekitClient.RoomEvent.ParticipantDisconnected, (participant) => {
      log("Participant left: " + participant.identity, "warn");
    });
    room.on(LivekitClient.RoomEvent.TrackSubscribed, (track, publication, participant) => {
      log("Track subscribed: " + track.kind + " from " + participant.identity);
      if (track.kind === LivekitClient.Track.Kind.Audio) {
        const element = track.attach();
        document.body.appendChild(element);
        element.play().then(() => {
          log("Audio playback started");
        }).catch((err) => {
          log("Audio play failed: " + err.message, "error");
        });
      }
    });
    room.on(LivekitClient.RoomEvent.TrackPublished, (publication, participant) => {
      log("Track published by " + participant.identity + ": " + publication.kind);
    });
    room.on(LivekitClient.RoomEvent.TrackUnsubscribed, (track, publication, participant) => {
      log("Track unsubscribed: " + track.kind + " from " + participant.identity, "warn");
    });
    room.on(LivekitClient.RoomEvent.DataReceived, (payload, participant) => {
      try {
        const msg = new TextDecoder().decode(payload);
        log("Data from " + (participant ? participant.identity : "server") + ": " + msg);
      } catch (e) {
        log("Data received (binary, " + payload.byteLength + " bytes)");
      }
    });
    room.on(LivekitClient.RoomEvent.ConnectionQualityChanged, (quality, participant) => {
      if (participant.isLocal) {
        log("Connection quality: " + quality);
      }
    });
    room.on(LivekitClient.RoomEvent.MediaDevicesError, (err) => {
      log("Media device error: " + err.message, "error");
    });
    room.on(LivekitClient.RoomEvent.RoomMetadataChanged, (metadata) => {
      log("Room metadata changed: " + metadata);
    });
    log("Connecting to " + LIVEKIT_URL + "...");
    await room.connect(data.url || LIVEKIT_URL, data.token);
    log("Enabling microphone...");
    await room.localParticipant.setMicrophoneEnabled(true);
    log("Microphone enabled. Speak now.");
    // Log local tracks
    room.localParticipant.audioTrackPublications.forEach((pub) => {
      log("Local audio track: " + pub.trackSid + " muted=" + pub.isMuted);
    });
  } catch (err) {
    setStatus("disconnected", "Error: " + err.message);
    log("Connection error: " + err.message, "error");
    console.error(err);
  }
 }
 async function disconnect() {
  if (room) {
    log("Disconnecting...");
    await room.disconnect();
    room = null;
  }
 }
 function setMode(mode) {
  document.getElementById("modeDisplay").textContent = "Mode: " + mode;
  log("Mode changed to: " + mode);
  if (room && room.localParticipant) {
    const encoder = new TextEncoder();
    room.localParticipant.publishData(
      encoder.encode(JSON.stringify({ type: "mode_change", mode })),
      { reliable: true }
    );
  }
 }
 function setStatus(state, text) {
  const el = document.getElementById("status");
  el.className = "status " + state;
  el.textContent = text;
 }
 document.getElementById("connectBtn").addEventListener("click", connect);
 document.getElementById("disconnectBtn").addEventListener("click", disconnect);
 log("VoiceCoach client loaded. Ready to connect.");
@@ -0,0 +1,46 @@
 <!DOCTYPE html>
 <html lang="en">
 <head>
  <meta charset="UTF-8">
  <meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0">
  <title>VoiceCoach</title>
  <link rel="stylesheet" href="style.css">
 </head>
 <body>
  <div class="container">
    <div class="header">
      <h1>VoiceCoach</h1>
      <div id="status" class="status disconnected">Disconnected</div>
    </div>
    <div class="controls">
      <button id="connectBtn" class="btn btn-primary">Start Session</button>
      <button id="disconnectBtn" class="btn btn-danger" style="display:none">End Session</button>
    </div>
    <div id="modeDisplay" class="mode-display">Mode: coaching</div>
    <div id="transcript" class="transcript">
      <p class="hint">Press "Start Session" to connect your voice.</p>
    </div>
    <div class="mode-buttons">
      <button class="mode-btn" onclick="setMode('coaching mode')">Coaching</button>
      <button class="mode-btn" onclick="setMode('vent mode')">Vent</button>
      <button class="mode-btn" onclick="setMode('practice mode')">Practice</button>
      <button class="mode-btn" onclick="setMode('briefing mode')">Briefing</button>
    </div>
    <div class="log-panel">
      <div class="log-header">
        <span>Log</span>
        <button class="log-clear" onclick="clearLog()">Clear</button>
      </div>
      <div id="logOutput" class="log-output"></div>
    </div>
  </div>
  <script src="https://cdn.jsdelivr.net/npm/livekit-client/dist/livekit-client.umd.min.js"></script>
  <script src="app.js"></script>
 </body>
 </html>
@@ -0,0 +1,15 @@
 server {
    listen 80;
    root /usr/share/nginx/html;
    index index.html;
    location /api/ {
        proxy_pass http://pipecat-agent:8080/;
        proxy_http_version 1.1;
        proxy_set_header Host $host;
    }
    location / {
        try_files $uri $uri/ /index.html;
    }
 }
@@ -0,0 +1,131 @@
 * { margin: 0; padding: 0; box-sizing: border-box; }
 body {
  font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
  background: #1a1a2e;
  color: #eee;
  min-height: 100vh;
  display: flex;
  justify-content: center;
  align-items: center;
 }
 .container {
  max-width: 420px;
  width: 100%;
  padding: 24px;
 }
 .header {
  display: flex;
  justify-content: space-between;
  align-items: center;
  margin-bottom: 24px;
 }
 h1 { font-size: 24px; font-weight: 600; }
 .status {
  padding: 4px 12px;
  border-radius: 12px;
  font-size: 12px;
  font-weight: 600;
  text-transform: uppercase;
 }
 .status.disconnected { background: #444; color: #aaa; }
 .status.connecting { background: #553; color: #ffa; }
 .status.connected { background: #253; color: #5f5; }
 .controls { margin-bottom: 20px; }
 .btn {
  width: 100%;
  padding: 14px;
  border: none;
  border-radius: 8px;
  font-size: 16px;
  font-weight: 600;
  cursor: pointer;
  transition: opacity 0.2s;
 }
 .btn:hover { opacity: 0.85; }
 .btn-primary { background: #4a6cf7; color: #fff; }
 .btn-danger { background: #f74a4a; color: #fff; }
 .mode-display {
  text-align: center;
  font-size: 13px;
  color: #888;
  margin-bottom: 16px;
 }
 .transcript {
  background: #16213e;
  border-radius: 8px;
  padding: 16px;
  min-height: 200px;
  max-height: 400px;
  overflow-y: auto;
  margin-bottom: 16px;
 }
 .transcript .hint { color: #666; font-style: italic; }
 .mode-buttons {
  display: grid;
  grid-template-columns: 1fr 1fr;
  gap: 8px;
 }
 .mode-btn {
  padding: 10px;
  background: #16213e;
  border: 1px solid #333;
  border-radius: 6px;
  color: #ccc;
  font-size: 13px;
  cursor: pointer;
  transition: background 0.2s;
 }
 .mode-btn:hover { background: #1a2744; border-color: #4a6cf7; }
 .log-panel {
  margin-top: 16px;
  background: #0d1117;
  border: 1px solid #333;
  border-radius: 8px;
  overflow: hidden;
 }
 .log-header {
  display: flex;
  justify-content: space-between;
  align-items: center;
  padding: 6px 12px;
  background: #161b22;
  font-size: 12px;
  font-weight: 600;
  color: #888;
  text-transform: uppercase;
  letter-spacing: 0.5px;
 }
 .log-clear {
  background: none;
  border: 1px solid #444;
  border-radius: 4px;
  color: #888;
  font-size: 11px;
  padding: 2px 8px;
  cursor: pointer;
 }
 .log-clear:hover { border-color: #666; color: #ccc; }
 .log-output {
  padding: 8px 12px;
  max-height: 200px;
  overflow-y: auto;
  font-family: 'SF Mono', 'Fira Code', monospace;
  font-size: 11px;
  line-height: 1.6;
 }
 .log-output div { word-break: break-all; }
 .log-time { color: #555; margin-right: 6px; }
 .log-info { color: #8b949e; }
 .log-warn { color: #d29922; }
 .log-error { color: #f85149; }
@@ -0,0 +1,31 @@
 http:
  routers:
    friday-voice:
      rule: "Host(`coach.hobbs.farm`)"
      entryPoints:
        - websecure
      middlewares:
        - security-headers-allow-mic
      service: friday-voice
      tls:
        certResolver: letsencrypt
    friday-voice-http:
      rule: "Host(`coach.hobbs.farm`)"
      entryPoints:
        - web
      middlewares:
        - friday-voice-https-redirect
      service: friday-voice
  middlewares:
    friday-voice-https-redirect:
      redirectScheme:
        scheme: https
        permanent: true
  services:
    friday-voice:
      loadBalancer:
        servers:
          # friday-voice runs on the .150 Windows box (Docker Desktop), port 8320
          - url: "http://192.168.86.150:8320"
@@ -0,0 +1,29 @@
 http:
  routers:
    livekit:
      rule: "Host(`voice.hobbs.farm`)"
      entryPoints:
        - websecure
      service: livekit
      tls:
        certResolver: letsencrypt
    livekit-http:
      rule: "Host(`voice.hobbs.farm`)"
      entryPoints:
        - web
      middlewares:
        - livekit-https-redirect
      service: livekit
  middlewares:
    livekit-https-redirect:
      redirectScheme:
        scheme: https
        permanent: true
  services:
    livekit:
      loadBalancer:
        servers:
          - url: "http://192.168.86.150:7880"
        passHostHeader: true
@@ -0,0 +1,30 @@
 http:
  routers:
    voicecoach:
      rule: "Host(`coach.hobbs.farm`)"
      entryPoints:
        - websecure
      middlewares:
        - security-headers-allow-mic
      service: voicecoach
      tls:
        certResolver: letsencrypt
    voicecoach-http:
      rule: "Host(`coach.hobbs.farm`)"
      entryPoints:
        - web
      middlewares:
        - voicecoach-https-redirect
      service: voicecoach
  middlewares:
    voicecoach-https-redirect:
      redirectScheme:
        scheme: https
        permanent: true
  services:
    voicecoach:
      loadBalancer:
        servers:
          - url: "http://192.168.86.150:8085"