Basic Pipecat Voice Pipeline

Learn how to trace a Pipecat voice pipeline with Noveum Trace

This guide shows a complete Pipecat voice pipeline (STT + LLM function calling + TTS) instrumented with Noveum Trace. You will see per-turn spans for STT, LLM, and TTS, plus (optionally) uploaded audio for each span.

🎯 Use Case

Drive-Thru Voice Agent: A voice-powered ordering bot that:

Listens for customer speech (STT)
Uses an LLM with function calling to build an order
Speaks back responses (TTS)

🚀 Complete Working Example

import os
 
from dotenv import load_dotenv
from loguru import logger
from pipecat.adapters.schemas.function_schema import FunctionSchema
from pipecat.adapters.schemas.tools_schema import ToolsSchema
from pipecat.audio.vad.silero import SileroVADAnalyzer
from pipecat.frames.frames import (
    FunctionCallResultProperties,
    LLMRunFrame,
    TTSSpeakFrame,
)
from pipecat.pipeline.pipeline import Pipeline
from pipecat.pipeline.runner import PipelineRunner
from pipecat.pipeline.task import PipelineParams, PipelineTask
from pipecat.processors.aggregators.llm_context import LLMContext
from pipecat.processors.aggregators.llm_response_universal import (
    LLMContextAggregatorPair,
    LLMUserAggregatorParams,
)
from pipecat.runner.types import RunnerArguments
from pipecat.runner.utils import create_transport
from pipecat.services.cartesia.tts import CartesiaTTSService
from pipecat.services.deepgram.stt import DeepgramSTTService
from pipecat.services.llm_service import FunctionCallParams
from pipecat.services.openai.llm import OpenAILLMService
from pipecat.transports.base_transport import BaseTransport, TransportParams
from pipecat.transports.daily.transport import DailyParams
from pipecat.transports.websocket.fastapi import FastAPIWebsocketParams
 
import noveum_trace
from noveum_trace.integrations.pipecat import NoveumPipecatTracer
 
load_dotenv(override=True)
 
 
# Simple simulated order storage
current_order = {"items": [], "total": 0.0}
MENU = {
    "burger": 5.99,
    "fries": 2.99,
}
 
 
async def add_item_to_order(params: FunctionCallParams):
    item = params.arguments.get("item", "").lower().replace(" ", "_")
    quantity = params.arguments.get("quantity", 1)
 
    if item not in MENU:
        await params.result_callback(
            {"success": False, "message": f"Sorry, we don't have {item} on our menu."}
        )
        return
 
    price = MENU[item]
    current_order["items"].append(
        {"name": item, "quantity": quantity, "price": price}
    )
    current_order["total"] += price * quantity
 
    await params.result_callback(
        {
            "success": True,
            "item": item,
            "quantity": quantity,
            "price": price,
            "current_total": current_order["total"],
        }
    )
 
 
async def view_current_order(params: FunctionCallParams):
    items_summary = [
        {"name": i["name"], "quantity": i["quantity"], "price": i["price"]}
        for i in current_order["items"]
    ]
 
    await params.result_callback(
        {
            "success": True,
            "items": items_summary,
            "total": current_order["total"],
        }
    )
 
 
async def confirm_order(params: FunctionCallParams):
    order_total = current_order["total"]
    current_order["items"] = []
    current_order["total"] = 0.0
 
    await params.result_callback(
        {
            "success": True,
            "message": f"Order confirmed. Total was ${order_total:.2f}.",
        }
    )
 
 
transport_params = {
    "daily": lambda: DailyParams(audio_in_enabled=True, audio_out_enabled=True),
    "twilio": lambda: FastAPIWebsocketParams(audio_in_enabled=True, audio_out_enabled=True),
    "webrtc": lambda: TransportParams(audio_in_enabled=True, audio_out_enabled=True),
}
 
 
async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
    logger.info("Starting pipecat drive-thru bot")
 
    # Initialize Noveum Trace before the pipeline starts.
    # In production, call noveum_trace.init() once at module level
    # rather than inside a per-connection function.
    noveum_trace.init(
        api_key=os.getenv("NOVEUM_API_KEY"),
        project=os.getenv("NOVEUM_PROJECT", "pipecat-drive-thru"),
    )
 
    stt = DeepgramSTTService(api_key=os.getenv("DEEPGRAM_API_KEY"))
    tts = CartesiaTTSService(
        api_key=os.getenv("CARTESIA_API_KEY"),
        settings=CartesiaTTSService.Settings(
            voice="71a7ad14-091c-4e8e-a314-022ece01c121",
        ),
    )
    llm = OpenAILLMService(
        api_key=os.getenv("OPENAI_API_KEY"),
        settings=OpenAILLMService.Settings(
            system_instruction=(
                "You are a friendly drive-thru order taker. "
                "Use add_item_to_order to build the order, "
                "use view_current_order to check it, and "
                "use confirm_order when the customer is done."
            ),
        ),
    )
 
    # Register function handlers for the LLM
    llm.register_function("add_item_to_order", add_item_to_order)
    llm.register_function("view_current_order", view_current_order)
    llm.register_function("confirm_order", confirm_order)
 
    # Define function schemas for Pipecat's tool/function calling
    add_item_schema = FunctionSchema(
        name="add_item_to_order",
        description="Add an item to the customer's order.",
        properties={
            "item": {"type": "string", "description": "Menu item name (e.g., 'burger', 'fries')"},
            "quantity": {"type": "integer", "description": "Quantity (default 1)", "default": 1},
        },
        required=["item"],
    )
 
    view_order_schema = FunctionSchema(
        name="view_current_order",
        description="View the current order with all items and total price.",
        properties={},
        required=[],
    )
 
    confirm_order_schema = FunctionSchema(
        name="confirm_order",
        description="Confirm and finalize the customer's order.",
        properties={},
        required=[],
    )
 
    tools = ToolsSchema(
        standard_tools=[add_item_schema, view_order_schema, confirm_order_schema]
    )
 
    context = LLMContext(tools=tools)
    user_aggregator, assistant_aggregator = LLMContextAggregatorPair(
        context,
        user_params=LLMUserAggregatorParams(vad_analyzer=SileroVADAnalyzer()),
    )
 
    pipeline = Pipeline(
        [
            transport.input(),
            stt,
            user_aggregator,
            llm,
            tts,
            transport.output(),
            assistant_aggregator,
        ]
    )
 
    tracer = NoveumPipecatTracer(record_audio=True)
 
    # 1) Wrap the pipeline. With record_audio=True this auto-inserts an
    #    AudioBufferProcessor (full stereo conversation WAV: left = user,
    #    right = bot). You MUST use the return value.
    pipeline = tracer.observe_pipeline(pipeline)
 
    task = PipelineTask(
        pipeline,
        params=PipelineParams(enable_metrics=True, enable_usage_metrics=True),
    )
 
    # 2) Register handlers — adds the observer, wires turn tracking, taps the
    #    transport for raw audio, and stamps session metadata. Use the return value.
    task = await tracer.register_task_handlers(task, transport=transport)
 
    @transport.event_handler("on_client_connected")
    async def on_client_connected(transport, client):
        logger.info("Customer connected to drive-thru")
        context.add_message({"role": "user", "content": "Greet the customer and take their order."})
        await task.queue_frames([LLMRunFrame()])
 
    @transport.event_handler("on_client_disconnected")
    async def on_client_disconnected(transport, client):
        logger.info("Customer disconnected")
        await task.cancel()
 
    runner = PipelineRunner(handle_sigint=runner_args.handle_sigint)
    await runner.run(task)
 
 
async def bot(runner_args: RunnerArguments):
    transport = await create_transport(runner_args, transport_params)
    await run_bot(transport, runner_args)
 
 
if __name__ == "__main__":
    from pipecat.runner.run import main
 
    main()

📋 Prerequisites

pip install "noveum-trace[pipecat]"

Set your environment variables:

export NOVEUM_API_KEY="your-noveum-api-key"
export NOVEUM_PROJECT="pipecat-drive-thru"
export DEEPGRAM_API_KEY="your-deepgram-api-key"
export CARTESIA_API_KEY="your-cartesia-api-key"
export OPENAI_API_KEY="your-openai-api-key"

🔧 How It Works

noveum_trace.init() configures where traces go. In production, call this once at module level rather than inside a per-connection function.
tracer = NoveumPipecatTracer(record_audio=True) sets what the tracer captures.
pipeline = tracer.observe_pipeline(pipeline) instruments the pipeline. With record_audio=True it auto-inserts an AudioBufferProcessor(num_channels=2) at the tail, so a stereo WAV is uploaded as a pipecat.full_conversation span when the session ends — no manual wiring. Use the return value.
task = await tracer.register_task_handlers(task, transport=transport) adds the observer, wires turn boundaries (so you get per-turn STT/LLM/TTS spans), taps the transport for raw input audio, and forces the metrics flags on. Use the return value.

📊 What You'll See in the Dashboard

pipecat.conversation trace with one pipecat.turn per user→bot exchange
Child spans per turn:
- pipecat.stt (recognized speech/text, model, confidence, latency)
- pipecat.llm (prompt, output, token usage, cost, function calls when emitted)
- pipecat.tts (spoken text, voice, audio timing)
When record_audio=True, per-span audio uploads add stt.audio_uuid and tts.audio_uuid attributes
A pipecat.full_conversation span at the trace root containing the full stereo WAV of the session (left channel = user, right channel = bot) — enabled by record_audio=True, which auto-inserts the AudioBufferProcessor for you

🔍 Troubleshooting

If turns don’t split correctly, verify that:

You assigned the return values of both tracer.observe_pipeline(...) and await tracer.register_task_handlers(...)
noveum_trace.init() is called before the pipeline starts

🚀 Next Steps

Read the integration overview: Pipecat Integration Overview