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:
+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)
|
||||
|
||||
Reference in New Issue
Block a user