Pipecat 통합

import asyncio
import os
from typing import AsyncIterator
from dotenv import load_dotenv
from pipecat.frames.frames import (
    Frame,
    LLMContextFrame,
    LLMFullResponseEndFrame,
    LLMTextFrame,
)
from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.runner import PipelineRunner
from pipecat.pipeline.task import PipelineTask
from pipecat.processors.aggregators.llm_context import LLMContext
from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
from pipecat.services.openai.llm import OpenAILLMService
load_dotenv()
SYSTEM_PROMPT = (
    "You are a helpful voice assistant. Keep responses concise and conversational."
)
class TextSink(FrameProcessor):
    """Drain LLMTextFrame text into an asyncio.Queue."""
    def __init__(self, queue: asyncio.Queue):
        super().__init__()
        self._queue = queue
    async def process_frame(self, frame: Frame, direction: FrameDirection):
        await super().process_frame(frame, direction)
        if isinstance(frame, LLMTextFrame):
            await self._queue.put(frame.text)
        elif isinstance(frame, LLMFullResponseEndFrame):
            await self._queue.put(None)  # sentinel
        await self.push_frame(frame, direction)
def build_messages(transcript: list[dict]) -> list[dict]:
    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
    for turn in transcript:
        role = "assistant" if turn["role"] == "agent" else turn["role"]
        messages.append({"role": role, "content": turn["content"]})
    return messages
async def run_pipecat_brain(transcript: list[dict]) -> AsyncIterator[str]:
    """Yield response text chunks from a one-turn Pipecat pipeline."""
    llm = OpenAILLMService(
        api_key=os.environ["OPENAI_API_KEY"],
        model="gpt-4o-mini",
    )
    queue: asyncio.Queue[str | None] = asyncio.Queue()
    sink = TextSink(queue)
    task = PipelineTask(Pipeline([llm, sink]))
    runner = PipelineRunner(handle_sigint=False)
    async def drive():
        context = LLMContext(build_messages(transcript))
        await task.queue_frame(LLMContextFrame(context))
        await task.stop_when_done()
    run_task = asyncio.create_task(runner.run(task))
    drive_task = asyncio.create_task(drive())
    try:
        while True:
            chunk = await queue.get()
            if chunk is None:
                break
            yield chunk
    finally:
        await task.cancel()
        await asyncio.gather(run_task, drive_task, return_exceptions=True)

import asyncio
import os
from dotenv import load_dotenv
from elevenlabs import AsyncElevenLabs
from brain import run_pipecat_brain
load_dotenv()
elevenlabs = AsyncElevenLabs(api_key=os.environ["ELEVENLABS_API_KEY"])
SPEECH_ENGINE_ID = os.environ["SPEECH_ENGINE_ID"]
async def on_transcript(transcript, session):
    history = [{"role": m.role, "content": m.content} for m in transcript]
    await session.send_response(run_pipecat_brain(history))
async def main():
    engine = await elevenlabs.speech_engine.get(SPEECH_ENGINE_ID)
    await engine.serve(
        port=3001,
        path="/ws",
        debug=True,
        on_transcript=on_transcript,
    )
if __name__ == "__main__":
    asyncio.run(main())