livekit-python-agents-examples/docs/examples/playing_audio at main

title

Playing Audio

Prerequisites

Add a .env in this directory with your LiveKit credentials:

LIVEKIT_URL=your_livekit_url
LIVEKIT_API_KEY=your_api_key
LIVEKIT_API_SECRET=your_api_secret

Install dependencies:

pip install "livekit-agents[silero]" python-dotenv

Place an audio.wav file in the same directory as the script

Load environment, logging, and define an AgentServer

Load environment variables, configure logging, and initialize the AgentServer.

import logging
from pathlib import Path
import wave
from dotenv import load_dotenv
from livekit.agents import JobContext, JobProcess, AgentServer, cli, Agent, AgentSession, inference, RunContext, function_tool
from livekit.plugins import silero
from livekit import rtc

load_dotenv()

logger = logging.getLogger("playing-audio")
logger.setLevel(logging.INFO)

server = AgentServer()

Define the agent with audio playback tool

Create a lightweight agent with instructions and a function tool that reads a WAV file, builds an AudioFrame, and streams it to the user.

class AudioPlayerAgent(Agent):
    def __init__(self) -> None:
        super().__init__(
            instructions="""
                You are a helpful assistant communicating through voice. Don't use any unpronouncable characters.
                If asked to play audio, use the `play_audio_file` function.
            """
        )

    @function_tool
    async def play_audio_file(self, context: RunContext):
        """Play a local audio file"""
        audio_path = Path(__file__).parent / "audio.wav"

        with wave.open(str(audio_path), 'rb') as wav_file:
            num_channels = wav_file.getnchannels()
            sample_rate = wav_file.getframerate()
            frames = wav_file.readframes(wav_file.getnframes())

        audio_frame = rtc.AudioFrame(
            data=frames,
            sample_rate=sample_rate,
            num_channels=num_channels,
            samples_per_channel=wav_file.getnframes()
        )

        async def audio_generator():
            yield audio_frame

        await self.session.say("Playing audio file", audio=audio_generator())

        return None, "I've played the audio file for you."

    async def on_enter(self):
        self.session.generate_reply()

Prewarm VAD for faster connections

Preload the VAD model once per process to reduce connection latency.

def prewarm(proc: JobProcess):
    proc.userdata["vad"] = silero.VAD.load()

server.setup_fnc = prewarm

Define the rtc session entrypoint

Create the session with STT/LLM/TTS configuration and start the audio player agent.

@server.rtc_session()
async def entrypoint(ctx: JobContext):
    ctx.log_context_fields = {"room": ctx.room.name}

    session = AgentSession(
        stt=inference.STT(model="deepgram/nova-3-general"),
        llm=inference.LLM(model="openai/gpt-5-mini"),
        tts=inference.TTS(model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
        vad=ctx.proc.userdata["vad"],
        preemptive_generation=True,
    )

    await session.start(agent=AudioPlayerAgent(), room=ctx.room)
    await ctx.connect()

Run the server

Start the agent server with the CLI runner.

if __name__ == "__main__":
    cli.run_app(server)

Run it

python playing_audio.py console

How it works

The agent greets the user on entry.
The LLM can invoke play_audio_file when asked to play audio.
The tool reads a local WAV file, wraps it in an AudioFrame, and streams it via session.say.
A short spoken preamble ("Playing audio file") plays before the audio clip.
The rest of the media pipeline continues unchanged.

Full example

import logging
from pathlib import Path
import wave
from dotenv import load_dotenv
from livekit.agents import JobContext, JobProcess, AgentServer, cli, Agent, AgentSession, inference, RunContext, function_tool
from livekit.plugins import silero
from livekit import rtc

load_dotenv()

logger = logging.getLogger("playing-audio")
logger.setLevel(logging.INFO)

class AudioPlayerAgent(Agent):
    def __init__(self) -> None:
        super().__init__(
            instructions="""
                You are a helpful assistant communicating through voice. Don't use any unpronouncable characters.
                If asked to play audio, use the `play_audio_file` function.
            """
        )

    @function_tool
    async def play_audio_file(self, context: RunContext):
        """Play a local audio file"""
        audio_path = Path(__file__).parent / "audio.wav"

        with wave.open(str(audio_path), 'rb') as wav_file:
            num_channels = wav_file.getnchannels()
            sample_rate = wav_file.getframerate()
            frames = wav_file.readframes(wav_file.getnframes())

        audio_frame = rtc.AudioFrame(
            data=frames,
            sample_rate=sample_rate,
            num_channels=num_channels,
            samples_per_channel=wav_file.getnframes()
        )

        async def audio_generator():
            yield audio_frame

        await self.session.say("Playing audio file", audio=audio_generator())

        return None, "I've played the audio file for you."

    async def on_enter(self):
        self.session.generate_reply()

server = AgentServer()

def prewarm(proc: JobProcess):
    proc.userdata["vad"] = silero.VAD.load()

server.setup_fnc = prewarm

@server.rtc_session()
async def entrypoint(ctx: JobContext):
    ctx.log_context_fields = {"room": ctx.room.name}

    session = AgentSession(
        stt=inference.STT(model="deepgram/nova-3-general"),
        llm=inference.LLM(model="openai/gpt-5-mini"),
        tts=inference.TTS(model="cartesia/sonic-3", voice="9626c31c-bec5-4cca-baa8-f8ba9e84c8bc"),
        vad=ctx.proc.userdata["vad"],
        preemptive_generation=True,
    )

    await session.start(agent=AudioPlayerAgent(), room=ctx.room)
    await ctx.connect()

if __name__ == "__main__":
    cli.run_app(server)