feat: VoiceCoach v2 - OpenClaw gateway + Speaches STT/TTS
- pipeline.py: completely rewritten to use Speaches STT/TTS and OpenClaw LLM - main.py: v2 with OPENCLAW_TOKEN support - requirements.txt: slimmed (removed anthropic, faster-whisper direct deps) - Dockerfile: slimmer build (no model pre-download) - whisper_stt.py, piper_tts.py: removed (replaced by Speaches API) - docker-compose.yml: v2 with speaches + openclaw services, piper-tts removed - openclaw-agent/system_prompt.md: new directory with agent system prompt
This commit is contained in:
+66
-18
@@ -1,6 +1,6 @@
|
||||
services:
|
||||
|
||||
# LiveKit WebRTC Server
|
||||
# LiveKit WebRTC Server (unchanged)
|
||||
livekit-server:
|
||||
image: livekit/livekit-server:latest
|
||||
container_name: voicecoach-livekit
|
||||
@@ -27,7 +27,39 @@ services:
|
||||
max-size: "50m"
|
||||
max-file: "3"
|
||||
|
||||
# SpeechBrain Emotion Analysis (internal only)
|
||||
# Speaches: self-hosted Whisper STT + Kokoro TTS (v2 - replaces Piper + in-process Whisper)
|
||||
speaches:
|
||||
image: ghcr.io/speaches-ai/speaches:latest-cpu
|
||||
container_name: voicecoach-speaches
|
||||
restart: unless-stopped
|
||||
environment:
|
||||
- WHISPER__MODEL=Systran/faster-whisper-large-v3-turbo
|
||||
- WHISPER__INFERENCE_DEVICE=cpu
|
||||
- WHISPER__COMPUTE_TYPE=int8
|
||||
- TTS__MODEL=kokoro
|
||||
- TTS__DEFAULT_VOICE=af_nova
|
||||
- TTS__LANGUAGE=en-us
|
||||
volumes:
|
||||
- /mnt/docker-data/configs/voicecoach/speaches-models:/root/.cache
|
||||
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
|
||||
@@ -36,6 +68,7 @@ services:
|
||||
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
|
||||
networks:
|
||||
@@ -50,18 +83,26 @@ services:
|
||||
max-size: "50m"
|
||||
max-file: "3"
|
||||
|
||||
# Piper TTS (internal only)
|
||||
piper-tts:
|
||||
build:
|
||||
context: ./piper-service
|
||||
image: voicecoach-piper:latest
|
||||
container_name: voicecoach-piper
|
||||
# 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"
|
||||
environment:
|
||||
- PIPER_MODEL=/app/models/en_US-lessac-medium.onnx
|
||||
- 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"]
|
||||
networks:
|
||||
- voicecoach
|
||||
mem_limit: 512M
|
||||
mem_limit: 2G
|
||||
memswap_limit: 2G
|
||||
labels:
|
||||
- "com.centurylinklabs.watchtower.enable=false"
|
||||
logging:
|
||||
@@ -70,7 +111,7 @@ services:
|
||||
max-size: "50m"
|
||||
max-file: "3"
|
||||
|
||||
# Pipecat Voice Pipeline Agent
|
||||
# Pipecat Voice Pipeline Agent (v2 - uses Speaches + OpenClaw gateway)
|
||||
pipecat-agent:
|
||||
build:
|
||||
context: ./pipecat-agent
|
||||
@@ -81,18 +122,25 @@ services:
|
||||
- .env
|
||||
environment:
|
||||
- LIVEKIT_URL=ws://livekit-server:7880
|
||||
- SPEECHBRAIN_SERVICE_URL=http://speechbrain-service:8001
|
||||
- PIPER_SERVICE_URL=http://piper-tts:8003
|
||||
- WHISPER_MODEL=medium
|
||||
- LIVEKIT_PUBLIC_URL=wss://voice.hobbs.farm
|
||||
- SPEACHES_URL=http://speaches:8000
|
||||
- SPEECHBRAIN_URL=http://speechbrain-service:8001
|
||||
- OPENCLAW_URL=http://voicecoach-openclaw:3100
|
||||
- OPENCLAW_AGENT_ID=voicecoach
|
||||
- WHISPER_MODEL=Systran/faster-whisper-large-v3-turbo
|
||||
- KOKORO_VOICE=af_nova
|
||||
- CLIPS_DIR=/data/clips/pending
|
||||
volumes:
|
||||
- /mnt/docker-data/configs/voicecoach/clips:/data/clips
|
||||
depends_on:
|
||||
- livekit-server
|
||||
- speaches
|
||||
- speechbrain-service
|
||||
- piper-tts
|
||||
- voicecoach-openclaw
|
||||
networks:
|
||||
- voicecoach
|
||||
mem_limit: 3G
|
||||
memswap_limit: 3G
|
||||
mem_limit: 1G
|
||||
memswap_limit: 1G
|
||||
labels:
|
||||
- "com.centurylinklabs.watchtower.enable=false"
|
||||
logging:
|
||||
@@ -101,7 +149,7 @@ services:
|
||||
max-size: "50m"
|
||||
max-file: "3"
|
||||
|
||||
# Web Client (phone interface)
|
||||
# Web Client (unchanged)
|
||||
voicecoach-web:
|
||||
build:
|
||||
context: ./web-client
|
||||
|
||||
Executable
+63
@@ -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.
|
||||
@@ -1,24 +1,14 @@
|
||||
FROM python:3.11-slim
|
||||
|
||||
RUN apt-get update && apt-get install -y \
|
||||
libsndfile1 \
|
||||
ffmpeg \
|
||||
&& rm -rf /var/lib/apt/lists/*
|
||||
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
|
||||
|
||||
# Pre-download faster-whisper medium model at build time
|
||||
RUN python -c "\
|
||||
from faster_whisper import WhisperModel; \
|
||||
WhisperModel('medium', device='cpu', compute_type='int8'); \
|
||||
print('Whisper model downloaded.')"
|
||||
|
||||
COPY . .
|
||||
|
||||
HEALTHCHECK --interval=30s --timeout=10s --start-period=30s \
|
||||
CMD python -c "print('pipeline process alive')" || exit 1
|
||||
CMD python -c "import httpx; httpx.get('http://localhost:8080/health').raise_for_status()"
|
||||
|
||||
CMD ["python", "main.py"]
|
||||
|
||||
+26
-29
@@ -1,11 +1,13 @@
|
||||
"""
|
||||
Entry point for the Pipecat VoiceCoach agent.
|
||||
Runs both the token HTTP server (for web clients) and the voice pipeline.
|
||||
VoiceCoach pipeline entry point -- v2.
|
||||
Runs token server + pipeline with unique identity per session.
|
||||
"""
|
||||
import asyncio
|
||||
import logging
|
||||
import os
|
||||
import threading
|
||||
import uuid
|
||||
from pathlib import Path
|
||||
|
||||
import structlog
|
||||
import uvicorn
|
||||
@@ -22,8 +24,9 @@ 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 (serves /token for web clients) ---
|
||||
# --- Token HTTP server ---
|
||||
|
||||
token_app = FastAPI(title="VoiceCoach Token Service")
|
||||
|
||||
@@ -39,17 +42,15 @@ class TokenResponse(BaseModel):
|
||||
|
||||
@token_app.get("/health")
|
||||
async def health():
|
||||
return {"status": "ok"}
|
||||
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,
|
||||
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)
|
||||
@@ -59,35 +60,31 @@ def run_token_server():
|
||||
uvicorn.run(token_app, host="0.0.0.0", port=8080, log_level="info")
|
||||
|
||||
|
||||
# --- Main pipeline ---
|
||||
# --- Main ---
|
||||
|
||||
async def main():
|
||||
anthropic_key = os.environ["ANTHROPIC_API_KEY"]
|
||||
|
||||
# Start token server in a background thread
|
||||
server_thread = threading.Thread(target=run_token_server, daemon=True)
|
||||
server_thread.start()
|
||||
logger.info("Token server started on :8080")
|
||||
|
||||
# Generate agent's own LiveKit token
|
||||
token = livekit_api.AccessToken(LIVEKIT_API_KEY, LIVEKIT_API_SECRET)
|
||||
token.with_grants(livekit_api.VideoGrants(room_join=True, room="voicecoach"))
|
||||
token.identity = "voicecoach-agent"
|
||||
livekit_token = token.to_jwt()
|
||||
while True:
|
||||
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
|
||||
livekit_token = token.to_jwt()
|
||||
|
||||
session_context = {
|
||||
"baseline_context": "",
|
||||
"session_history": "",
|
||||
"current_mode": "coaching mode",
|
||||
}
|
||||
logger.info("Starting VoiceCoach v2 pipeline", identity=agent_identity)
|
||||
try:
|
||||
await run_voice_coach_pipeline(
|
||||
livekit_url=LIVEKIT_WS_URL,
|
||||
livekit_token=livekit_token,
|
||||
openclaw_token=OPENCLAW_TOKEN,
|
||||
)
|
||||
except Exception as e:
|
||||
logger.error("Pipeline error, restarting in 35s", error=str(e))
|
||||
|
||||
logger.info("Starting VoiceCoach pipeline", livekit_url=LIVEKIT_WS_URL)
|
||||
await run_voice_coach_pipeline(
|
||||
livekit_url=LIVEKIT_WS_URL,
|
||||
livekit_token=livekit_token,
|
||||
anthropic_key=anthropic_key,
|
||||
session_context=session_context,
|
||||
)
|
||||
await asyncio.sleep(35)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
+203
-129
@@ -1,157 +1,224 @@
|
||||
"""
|
||||
Main Pipecat pipeline assembly.
|
||||
Connects LiveKit -> Whisper STT -> SpeechBrain -> Claude -> Piper TTS -> LiveKit.
|
||||
Fully self-hosted STT/TTS (no external API dependencies except Anthropic).
|
||||
Pipecat pipeline -- v2 fully self-hosted.
|
||||
|
||||
STT: Speaches (faster-whisper) via OpenAI-compatible /v1/audio/transcriptions
|
||||
LLM: OpenClaw VoiceCoach agent via OpenAI-compatible gateway
|
||||
TTS: Speaches (Kokoro 82M) via OpenAI-compatible /v1/audio/speech
|
||||
"""
|
||||
import asyncio
|
||||
import io
|
||||
import os
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
import uuid
|
||||
import wave
|
||||
import httpx
|
||||
from typing import Optional
|
||||
from datetime import datetime
|
||||
from pathlib import Path
|
||||
from typing import AsyncGenerator, Optional
|
||||
|
||||
import httpx
|
||||
from pipecat.frames.frames import (
|
||||
AudioRawFrame, TranscriptionFrame, TextFrame, EndFrame, TTSAudioRawFrame
|
||||
AudioRawFrame, EndFrame, ErrorFrame, Frame, TextFrame, TranscriptionFrame,
|
||||
UserStartedSpeakingFrame, UserStoppedSpeakingFrame,
|
||||
)
|
||||
from pipecat.pipeline.pipeline import Pipeline
|
||||
from pipecat.pipeline.runner import PipelineRunner
|
||||
from pipecat.pipeline.task import PipelineParams, PipelineTask
|
||||
from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
|
||||
from pipecat.services.anthropic import AnthropicLLMService
|
||||
from pipecat.transports.services.livekit import LiveKitTransport, LiveKitParams
|
||||
from pipecat.processors.frame_processor import FrameProcessor, FrameDirection
|
||||
from pipecat.vad.silero import SileroVADAnalyzer
|
||||
from pipecat.services.ai_services import STTService
|
||||
from pipecat.services.openai import OpenAILLMService, OpenAITTSService
|
||||
from pipecat.transports.services.livekit import LiveKitTransport, LiveKitParams
|
||||
|
||||
from prompt_builder import build_system_prompt, build_user_message, VocalState
|
||||
from whisper_stt import LocalWhisperSTT
|
||||
from piper_tts import PiperTTSClient
|
||||
try:
|
||||
from pipecat.audio.vad.silero import SileroVADAnalyzer
|
||||
from pipecat.audio.vad.vad_analyzer import VADParams
|
||||
except ImportError:
|
||||
from pipecat.vad.silero import SileroVADAnalyzer
|
||||
from pipecat.vad.vad_analyzer import VADParams
|
||||
|
||||
from prompt_builder import build_user_message, VocalState
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
SPEECHBRAIN_URL = os.getenv("SPEECHBRAIN_SERVICE_URL", "http://speechbrain-service:8001")
|
||||
PIPER_URL = os.getenv("PIPER_SERVICE_URL", "http://piper-tts:8003")
|
||||
ANTHROPIC_MODEL = "claude-sonnet-4-6"
|
||||
WHISPER_MODEL = os.getenv("WHISPER_MODEL", "medium")
|
||||
SPEACHES_URL = os.getenv("SPEACHES_URL", "http://speaches:8000")
|
||||
OPENCLAW_URL = os.getenv("OPENCLAW_URL", "http://voicecoach-openclaw:3100")
|
||||
SPEECHBRAIN_URL = os.getenv("SPEECHBRAIN_URL", "http://speechbrain-service:8001")
|
||||
|
||||
WHISPER_MODEL = os.getenv("WHISPER_MODEL", "Systran/faster-whisper-large-v3-turbo")
|
||||
KOKORO_VOICE = os.getenv("KOKORO_VOICE", "af_nova")
|
||||
OPENCLAW_AGENT_ID = os.getenv("OPENCLAW_AGENT_ID", "voicecoach")
|
||||
|
||||
CLIPS_DIR = Path(os.getenv("CLIPS_DIR", "/data/clips/pending"))
|
||||
UNCERTAINTY_THRESHOLD = 0.60
|
||||
MIN_CLIP_DURATION = 1.0
|
||||
|
||||
|
||||
class WhisperSTTProcessor(FrameProcessor):
|
||||
"""Buffers audio from VAD and transcribes with faster-whisper on end-of-speech."""
|
||||
class SpeachesSTTService(STTService):
|
||||
"""STT via Speaches OpenAI-compatible /v1/audio/transcriptions endpoint.
|
||||
|
||||
def __init__(self, whisper: LocalWhisperSTT, **kwargs):
|
||||
pipecat-ai 0.0.48 does not include OpenAISTTService; this thin wrapper
|
||||
calls the Speaches HTTP API directly using the STTService base class.
|
||||
"""
|
||||
|
||||
def __init__(self, base_url: str, model: str, language: str = "en", **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
self._whisper = whisper
|
||||
self._audio_buffer: list[bytes] = []
|
||||
self._base_url = base_url.rstrip("/")
|
||||
self._model = model
|
||||
self._language = language
|
||||
self._http = httpx.AsyncClient(timeout=30.0)
|
||||
|
||||
async def process_frame(self, frame, direction: FrameDirection):
|
||||
await super().process_frame(frame, direction)
|
||||
async def set_model(self, model: str):
|
||||
self._model = model
|
||||
|
||||
if isinstance(frame, AudioRawFrame):
|
||||
self._audio_buffer.append(frame.audio)
|
||||
elif isinstance(frame, EndFrame):
|
||||
# VAD detected end of speech -- transcribe
|
||||
if self._audio_buffer:
|
||||
audio_bytes = b"".join(self._audio_buffer)
|
||||
self._audio_buffer = []
|
||||
text = self._whisper.transcribe(audio_bytes)
|
||||
if text.strip():
|
||||
await self.push_frame(
|
||||
TranscriptionFrame(text=text, user_id="jeremy", timestamp=""),
|
||||
direction,
|
||||
)
|
||||
await self.push_frame(frame, direction)
|
||||
else:
|
||||
await self.push_frame(frame, direction)
|
||||
async def set_language(self, language):
|
||||
self._language = str(language)
|
||||
|
||||
async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
|
||||
if not audio:
|
||||
return
|
||||
# Wrap raw PCM in a WAV container for the API
|
||||
buf = io.BytesIO()
|
||||
with wave.open(buf, "wb") as wf:
|
||||
wf.setnchannels(1)
|
||||
wf.setsampwidth(2)
|
||||
wf.setframerate(16000)
|
||||
wf.writeframes(audio)
|
||||
wav_bytes = buf.getvalue()
|
||||
|
||||
try:
|
||||
r = await self._http.post(
|
||||
f"{self._base_url}/v1/audio/transcriptions",
|
||||
files={"file": ("audio.wav", wav_bytes, "audio/wav")},
|
||||
data={"model": self._model, "language": self._language, "response_format": "json"},
|
||||
)
|
||||
r.raise_for_status()
|
||||
text = r.json().get("text", "").strip()
|
||||
if text:
|
||||
logger.info("STT transcription: %s", text[:100])
|
||||
yield TranscriptionFrame(text=text, user_id="user", timestamp=datetime.utcnow().isoformat())
|
||||
except Exception as e:
|
||||
logger.warning("Speaches STT error: %s", e)
|
||||
yield ErrorFrame(error=str(e))
|
||||
|
||||
|
||||
class VocalStateInjector(FrameProcessor):
|
||||
"""Intercepts transcriptions, fetches vocal analysis from SpeechBrain,
|
||||
and enriches the frame before passing to the LLM."""
|
||||
"""Buffers audio per utterance, sends to SpeechBrain for analysis,
|
||||
enriches TranscriptionFrame with [VOCAL_STATE] metadata,
|
||||
and saves clips for active labeling."""
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
def __init__(self, session_id: str, **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
self._current_audio: list[bytes] = []
|
||||
self._http_client = httpx.AsyncClient(timeout=3.0)
|
||||
self._session_id = session_id
|
||||
self._audio_buffer: list[bytes] = []
|
||||
self._is_speaking = False
|
||||
self._http = httpx.AsyncClient(timeout=3.0)
|
||||
|
||||
async def process_frame(self, frame, direction: FrameDirection):
|
||||
await super().process_frame(frame, direction)
|
||||
|
||||
if isinstance(frame, AudioRawFrame):
|
||||
self._current_audio.append(frame.audio)
|
||||
if isinstance(frame, UserStartedSpeakingFrame):
|
||||
self._is_speaking = True
|
||||
self._audio_buffer = []
|
||||
logger.info("VAD: user started speaking")
|
||||
await self.push_frame(frame, direction)
|
||||
elif isinstance(frame, AudioRawFrame) and self._is_speaking:
|
||||
self._audio_buffer.append(frame.audio)
|
||||
await self.push_frame(frame, direction)
|
||||
elif isinstance(frame, UserStoppedSpeakingFrame):
|
||||
self._is_speaking = False
|
||||
logger.info("VAD: user stopped speaking, buffer: %d chunks", len(self._audio_buffer))
|
||||
await self.push_frame(frame, direction)
|
||||
elif isinstance(frame, TranscriptionFrame):
|
||||
vocal_state = await self._analyze_buffered_audio()
|
||||
self._current_audio = []
|
||||
wav_bytes = self._flush_audio_to_wav()
|
||||
self._audio_buffer = []
|
||||
|
||||
vocal_state = None
|
||||
if wav_bytes:
|
||||
vocal_state, analysis = await self._analyze(wav_bytes)
|
||||
if analysis and self._clip_worth_saving(analysis, wav_bytes):
|
||||
await self._save_clip(wav_bytes, analysis)
|
||||
|
||||
enriched_text = build_user_message(frame.text, vocal_state)
|
||||
enriched_frame = TranscriptionFrame(
|
||||
text=enriched_text, user_id=frame.user_id, timestamp=frame.timestamp,
|
||||
logger.info("Enriched transcription: %s", enriched_text[:200])
|
||||
await self.push_frame(
|
||||
TranscriptionFrame(text=enriched_text, user_id=frame.user_id, timestamp=frame.timestamp),
|
||||
direction,
|
||||
)
|
||||
await self.push_frame(enriched_frame, direction)
|
||||
else:
|
||||
await self.push_frame(frame, direction)
|
||||
|
||||
async def _analyze_buffered_audio(self) -> Optional[VocalState]:
|
||||
if not self._current_audio:
|
||||
def _flush_audio_to_wav(self) -> Optional[bytes]:
|
||||
if not self._audio_buffer:
|
||||
return None
|
||||
try:
|
||||
audio_bytes = b"".join(self._current_audio)
|
||||
buf = io.BytesIO()
|
||||
with wave.open(buf, "wb") as wf:
|
||||
wf.setnchannels(1)
|
||||
wf.setsampwidth(2)
|
||||
wf.setframerate(16000)
|
||||
wf.writeframes(audio_bytes)
|
||||
wav_bytes = buf.getvalue()
|
||||
raw = b"".join(self._audio_buffer)
|
||||
buf = io.BytesIO()
|
||||
with wave.open(buf, "wb") as wf:
|
||||
wf.setnchannels(1)
|
||||
wf.setsampwidth(2)
|
||||
wf.setframerate(16000)
|
||||
wf.writeframes(raw)
|
||||
return buf.getvalue()
|
||||
|
||||
response = await self._http_client.post(
|
||||
async def _analyze(self, wav_bytes: bytes):
|
||||
try:
|
||||
r = await self._http.post(
|
||||
f"{SPEECHBRAIN_URL}/analyze",
|
||||
files={"audio": ("utterance.wav", wav_bytes, "audio/wav")},
|
||||
files={"audio": ("u.wav", wav_bytes, "audio/wav")},
|
||||
)
|
||||
response.raise_for_status()
|
||||
data = response.json()
|
||||
return VocalState(
|
||||
emotion=data.get("emotion", "neutral"),
|
||||
confidence=data.get("emotion_confidence", 0.0),
|
||||
pitch_mean_hz=data.get("pitch_mean_hz"),
|
||||
energy_rms=data.get("energy_rms", 0.0),
|
||||
speaking_rate_approx=data.get("speaking_rate_approx", 0.0),
|
||||
voiced_fraction=data.get("voiced_fraction", 0.0),
|
||||
r.raise_for_status()
|
||||
d = r.json()
|
||||
return (
|
||||
VocalState(
|
||||
emotion=d.get("emotion", "neutral"),
|
||||
confidence=d.get("emotion_confidence", 0.0),
|
||||
pitch_mean_hz=d.get("pitch_mean_hz"),
|
||||
energy_rms=d.get("energy_rms", 0.0),
|
||||
speaking_rate_approx=d.get("speaking_rate_approx", 0.0),
|
||||
voiced_fraction=d.get("voiced_fraction", 0.0),
|
||||
),
|
||||
d,
|
||||
)
|
||||
except Exception as e:
|
||||
logger.warning(f"SpeechBrain analysis failed (continuing without): {e}")
|
||||
return None
|
||||
logger.warning("SpeechBrain unavailable: %s -- continuing without vocal state", e)
|
||||
return None, None
|
||||
|
||||
def _clip_worth_saving(self, analysis: dict, wav_bytes: bytes) -> bool:
|
||||
duration = len(wav_bytes) / (16000 * 2)
|
||||
return duration >= MIN_CLIP_DURATION
|
||||
|
||||
class PiperTTSProcessor(FrameProcessor):
|
||||
"""Converts text frames to audio via Piper TTS service."""
|
||||
|
||||
def __init__(self, piper_client: PiperTTSClient, **kwargs):
|
||||
super().__init__(**kwargs)
|
||||
self._piper = piper_client
|
||||
|
||||
async def process_frame(self, frame, direction: FrameDirection):
|
||||
await super().process_frame(frame, direction)
|
||||
|
||||
if isinstance(frame, TextFrame):
|
||||
wav_bytes = await self._piper.synthesize(frame.text)
|
||||
if wav_bytes:
|
||||
await self.push_frame(
|
||||
TTSAudioRawFrame(audio=wav_bytes, sample_rate=22050, num_channels=1),
|
||||
direction,
|
||||
)
|
||||
else:
|
||||
await self.push_frame(frame, direction)
|
||||
async def _save_clip(self, wav_bytes: bytes, analysis: dict):
|
||||
try:
|
||||
CLIPS_DIR.mkdir(parents=True, exist_ok=True)
|
||||
clip_id = f"{self._session_id}_{uuid.uuid4().hex[:8]}"
|
||||
(CLIPS_DIR / f"{clip_id}.wav").write_bytes(wav_bytes)
|
||||
meta = {
|
||||
"clip_id": clip_id,
|
||||
"session_id": self._session_id,
|
||||
"timestamp": datetime.utcnow().isoformat(),
|
||||
"duration_seconds": len(wav_bytes) / (16000 * 2),
|
||||
"speechbrain_emotion": analysis.get("emotion"),
|
||||
"speechbrain_confidence": analysis.get("emotion_confidence"),
|
||||
"emotion_scores": analysis.get("emotion_scores", {}),
|
||||
"pitch_mean_hz": analysis.get("pitch_mean_hz"),
|
||||
"energy_rms": analysis.get("energy_rms"),
|
||||
"speaking_rate_approx": analysis.get("speaking_rate_approx"),
|
||||
"labeled": False,
|
||||
"confirmed_label": None,
|
||||
"uncertain": analysis.get("emotion_confidence", 1.0) < UNCERTAINTY_THRESHOLD,
|
||||
}
|
||||
(CLIPS_DIR / f"{clip_id}.json").write_text(json.dumps(meta, indent=2))
|
||||
logger.info("Saved clip: %s", clip_id)
|
||||
except Exception as e:
|
||||
logger.warning("Failed to save clip: %s", e)
|
||||
|
||||
|
||||
async def run_voice_coach_pipeline(
|
||||
livekit_url: str,
|
||||
livekit_token: str,
|
||||
anthropic_key: str,
|
||||
session_context: dict,
|
||||
openclaw_token: str,
|
||||
):
|
||||
"""Assemble and run the full voice coaching pipeline."""
|
||||
|
||||
whisper = LocalWhisperSTT(model_size=WHISPER_MODEL, device="cpu", compute_type="int8")
|
||||
piper_client = PiperTTSClient(base_url=PIPER_URL)
|
||||
session_id = datetime.utcnow().strftime("%Y%m%d-%H%M%S")
|
||||
|
||||
transport = LiveKitTransport(
|
||||
url=livekit_url,
|
||||
@@ -160,39 +227,52 @@ async def run_voice_coach_pipeline(
|
||||
params=LiveKitParams(
|
||||
audio_in_enabled=True,
|
||||
audio_out_enabled=True,
|
||||
vad_analyzer=SileroVADAnalyzer(),
|
||||
vad_enabled=True,
|
||||
vad_analyzer=SileroVADAnalyzer(params=VADParams(
|
||||
confidence=0.5,
|
||||
start_secs=0.2,
|
||||
stop_secs=0.8,
|
||||
min_volume=0.1,
|
||||
)),
|
||||
vad_audio_passthrough=True,
|
||||
),
|
||||
)
|
||||
|
||||
whisper_stt = WhisperSTTProcessor(whisper)
|
||||
vocal_injector = VocalStateInjector()
|
||||
# STT: Speaches / faster-whisper via OpenAI-compatible API
|
||||
stt = SpeachesSTTService(
|
||||
base_url=SPEACHES_URL,
|
||||
model=WHISPER_MODEL,
|
||||
language="en",
|
||||
)
|
||||
|
||||
llm = AnthropicLLMService(
|
||||
api_key=anthropic_key,
|
||||
model=ANTHROPIC_MODEL,
|
||||
vocal_injector = VocalStateInjector(session_id=session_id)
|
||||
|
||||
# LLM: OpenClaw VoiceCoach agent via OpenAI-compatible gateway
|
||||
llm = OpenAILLMService(
|
||||
api_key=openclaw_token,
|
||||
base_url=f"{OPENCLAW_URL}/v1",
|
||||
model=f"openclaw/{OPENCLAW_AGENT_ID}",
|
||||
max_tokens=512,
|
||||
)
|
||||
|
||||
system_prompt = build_system_prompt(
|
||||
baseline_context=session_context.get("baseline_context", ""),
|
||||
session_history=session_context.get("session_history", ""),
|
||||
current_mode=session_context.get("current_mode", "coaching mode"),
|
||||
)
|
||||
|
||||
context = OpenAILLMContext(
|
||||
messages=[{"role": "system", "content": system_prompt}]
|
||||
)
|
||||
context = OpenAILLMContext(messages=[])
|
||||
context_aggregator = llm.create_context_aggregator(context)
|
||||
|
||||
piper_tts = PiperTTSProcessor(piper_client)
|
||||
# TTS: Speaches / Kokoro via OpenAI-compatible API
|
||||
tts = OpenAITTSService(
|
||||
api_key="speaches",
|
||||
base_url=f"{SPEACHES_URL}/v1",
|
||||
model="kokoro",
|
||||
voice=KOKORO_VOICE,
|
||||
)
|
||||
|
||||
pipeline = Pipeline([
|
||||
transport.input(),
|
||||
whisper_stt,
|
||||
stt,
|
||||
vocal_injector,
|
||||
context_aggregator.user(),
|
||||
llm,
|
||||
piper_tts,
|
||||
tts,
|
||||
transport.output(),
|
||||
context_aggregator.assistant(),
|
||||
])
|
||||
@@ -201,22 +281,16 @@ async def run_voice_coach_pipeline(
|
||||
pipeline,
|
||||
params=PipelineParams(allow_interruptions=True),
|
||||
)
|
||||
|
||||
runner = PipelineRunner()
|
||||
|
||||
@transport.event_handler("on_first_participant_joined")
|
||||
async def on_first_participant_joined(transport, participant):
|
||||
logger.info(f"Participant joined: {participant['identity']}")
|
||||
await task.queue_frames([
|
||||
TextFrame("VoiceCoach connected. How are you doing today?")
|
||||
])
|
||||
logger.info("Participant joined: %s", participant)
|
||||
await task.queue_frames([TextFrame("session_start")])
|
||||
|
||||
@transport.event_handler("on_participant_left")
|
||||
async def on_participant_left(transport, participant, reason):
|
||||
logger.info(f"Participant left: {participant['identity']}")
|
||||
logger.info("Session ending: %s", session_id)
|
||||
await task.queue_frames([EndFrame()])
|
||||
|
||||
try:
|
||||
await runner.run(task)
|
||||
finally:
|
||||
await piper_client.close()
|
||||
await runner.run(task)
|
||||
|
||||
@@ -1,34 +0,0 @@
|
||||
"""
|
||||
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()
|
||||
@@ -1,12 +1,7 @@
|
||||
pipecat-ai[livekit,silero,openai]==0.0.48
|
||||
fastapi==0.115.0
|
||||
uvicorn[standard]==0.30.0
|
||||
anthropic==0.34.0
|
||||
faster-whisper==1.0.3
|
||||
requests==2.32.3
|
||||
httpx==0.27.0
|
||||
python-dotenv==1.0.1
|
||||
structlog==24.4.0
|
||||
numpy==1.26.4
|
||||
soundfile==0.12.1
|
||||
livekit-api==0.7.1
|
||||
fastapi==0.115.0
|
||||
uvicorn[standard]==0.30.0
|
||||
|
||||
@@ -1,35 +0,0 @@
|
||||
"""
|
||||
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
|
||||
@@ -35,7 +35,7 @@ app = FastAPI(title="Piper TTS Service", lifespan=lifespan)
|
||||
|
||||
class TTSRequest(BaseModel):
|
||||
text: str
|
||||
speaker_id: int = 0
|
||||
speaker_id: int = None
|
||||
length_scale: float = 1.0
|
||||
sentence_silence: float = 0.2
|
||||
|
||||
@@ -57,13 +57,13 @@ async def synthesize(req: TTSRequest):
|
||||
|
||||
buf = io.BytesIO()
|
||||
with wave.open(buf, "wb") as wf:
|
||||
voice.synthesize(
|
||||
req.text,
|
||||
wf,
|
||||
speaker_id=req.speaker_id,
|
||||
length_scale=req.length_scale,
|
||||
sentence_silence=req.sentence_silence,
|
||||
)
|
||||
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(),
|
||||
|
||||
Reference in New Issue
Block a user