I’m having a problem. I created a very simple agen...
# ask-aid
delightful-hamburger-90968
12/20/2025, 9:54 PMI’m having a problem. I created a very simple agent running locally in console mode, and it works very well.
However, when I stream audio from the front end to LiveKit Cloud, the logs show that the local agent can hear the audio and generate responses, but no sound is played on the front end.
How should I debug this issue?
t
tall-airline-16756
12/20/2025, 9:54 PMThanks for reaching out, LUC. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
tall-airline-16756
12/20/2025, 9:55 PMThanks — I hear you: the agent hears and responds (logs show TTS), but the browser plays no sound. Below are focused debugging steps to find the root cause and fix it.
1) Confirm audio output is enabled on the agent
• Ensure you did not set audio output off (e.g.,
audio_enabled=Falsesession.say()<RoomAudioRenderer/>roomaudiorendereraudioTrackuseVoiceAssistant<audio>session.startsession.say<audio>d
delightful-hamburger-90968
12/20/2025, 9:58 PMBased on the following code, help me check it
import asyncio
import os
from dotenv import load_dotenv
from livekit import agents
from livekit.agents.voice import AgentSession, Agent
from livekit.plugins.silero import VAD
from livekit.plugins.deepgram import STT as DeepgramSTT, TTS as DeepgramTTS
from livekit.plugins.openai import LLM
from livekit import api
# Load environment
load_dotenv()
LIVEKIT_URL = os.getenv("LIVEKIT_URL")
LIVEKIT_API_KEY = os.getenv("LIVEKIT_API_KEY")
LIVEKIT_API_SECRET = os.getenv("LIVEKIT_API_SECRET")
DEEPGRAM_API_KEY = os.getenv("DEEPGRAM_API_KEY")
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
AGENT_NAME = ""
ROOM_NAME = "test_room"
USER_NAME = "test_user"
# -------------------------------
# Custom agent logic
# -------------------------------
class Assistant(Agent):
def __init__(self):
super().__init__(
id="voice_agent",
instructions="You are a helpful, concise English-speaking assistant."
)
# -------------------------------
# Main entrypoint
# -------------------------------
async def entrypoint(ctx: agents.JobContext):
# Connect to LiveKit Cloud
await ctx.connect(
# url=LIVEKIT_URL,
# api_key=LIVEKIT_API_KEY,
# api_secret=LIVEKIT_API_SECRET,
)
# Create AgentSession with your models
session = AgentSession(
stt=DeepgramSTT(
api_key=DEEPGRAM_API_KEY,
model="nova-2-general", # Deepgram STT model
language="en-US",
),
llm=LLM(
api_key=OPENAI_API_KEY,
model="gpt-4o-mini", # LLM model
),
tts=DeepgramTTS(
api_key=DEEPGRAM_API_KEY,
model="aura-asteria-en", # Deepgram TTS voice
),
vad=VAD.load(), # Silero VAD for turn detection
)
# Print LLM configuration
print(f"Session LLM: {session.llm}")
print(f"Session STT: {session.stt}")
print(f"Session TTS: {session.tts}")
print(f"Session VAD: {session.vad}")
@session.on("user_input_transcribed")
def on_user_input(ev):
print(f"[STT] Transcribed: '{ev.transcript}' (final={ev.is_final}, language={ev.language})")
@session.on("agent_state_changed")
def on_agent_state(ev):
print(f"[AGENT] State changed: {ev.old_state} -> {ev.new_state}")
@session.on("conversation_item_added")
def on_conversation_item(ev):
print(f"[CHAT] Item added: {ev.item}")
@session.on("speech_created")
def on_speech_created(ev):
print(f"[SPEECH] Created from '{ev.source}' (user_initiated={ev.user_initiated})")
@session.on("track_published")
def on_track_published(ev):
print(f"[TRACK] Published: kind={ev.track.kind}")
@session.on("error")
def on_error(ev):
print(f"[ERROR] {ev.error}")
@session.on("close")
def on_close(ev):
print(f"[CLOSE] Session closed. Reason: {ev.reason}, Error: {ev.error}")
# Start session
await session.start(
room=ctx.room,
agent=Assistant(),
)
print("RoomIO created:", session._room_io is not None)
print("Room IO:", session._room_io)
if session.output.audio is None:
print("❌ Audio output NOT attached")
else:
print("✅ Audio output attached:", type(session.output.audio))
# -------------------------------
# CLI launch
# -------------------------------
if name == "__main__":
agents.cli.run_app(
agents.WorkerOptions(
entrypoint_fnc=entrypoint,
port=8081,
)
)
Here is format of js
const { Room, RoomEvent } = window.LivekitClient;
let room; // keep global for debugging
let audioContext;
let analyser;
let checkAudioInterval;
function detectAudio(stream) {
const audioStatus = document.getElementById("audioStatus");
// Create audio context and analyser
audioContext = new (window.AudioContext || window.webkitAudioContext)();
analyser = audioContext.createAnalyser();
const source = audioContext.createMediaStreamSource(stream);
source.connect(analyser);
analyser.fftSize = 256;
const bufferLength = analyser.frequencyBinCount;
const dataArray = new Uint8Array(bufferLength);
// Check audio levels periodically
checkAudioInterval = setInterval(() => {
analyser.getByteFrequencyData(dataArray);
// Calculate average volume
const average = dataArray.reduce((a, b) => a + b) / bufferLength;
// Show "talking" if volume is above threshold
if (average > 10) {
audioStatus.classList.add("active");
} else {
audioStatus.classList.remove("active");
}
}, 100); // Check every 100ms
}
function stopAudioDetection() {
if (checkAudioInterval) {
clearInterval(checkAudioInterval);
checkAudioInterval = null;
}
if (audioContext) {
audioContext.close();
audioContext = null;
}
const audioStatus = document.getElementById("audioStatus");
audioStatus.classList.remove("active");
}
function attachAudioTrack(track, participant) {
if (track.kind !== "audio") return;
const audio = document.createElement("audio");
audio.autoplay = true;
audio.playsInline = true;
const stream = new MediaStream([track.mediaStreamTrack]);
audio.srcObject = stream;
document.body.appendChild(audio);
console.log("🔊 Audio from", participant.identity);
// Start audio detection
detectAudio(stream);
}
function toggleButtons(connected) {
const joinBtn = document.getElementById("joinBtn");
const disconnectBtn = document.getElementById("disconnectBtn");
if (connected) {
joinBtn.classList.add("hidden");
disconnectBtn.classList.remove("hidden");
} else {
joinBtn.classList.remove("hidden");
disconnectBtn.classList.add("hidden");
}
}
document.getElementById("joinBtn").onclick = async () => {
const tokenServerUrl = document.getElementById("tokenServerUrl").value.trim();
const roomName = document.getElementById("roomName").value.trim();
const userName = document.getElementById("userName").value.trim();
const status = document.getElementById("status");
if (!tokenServerUrl) {
status.innerText = "Missing Token Server URL";
return;
}
try {
// Fetch token from API
status.innerText = "Fetching token...";
console.log("🔑 Fetching token from server...");
const apiUrl = `${tokenServerUrl}/api/token?room=${encodeURIComponent(roomName)}&user=${encodeURIComponent(userName)}`;
const response = await fetch(apiUrl);
if (!response.ok) {
throw new Error(
Failed to fetch token: ${response.statusText}t
tall-airline-16756
12/20/2025, 9:58 PMThanks — this is helpful. I checked both snippets and the most likely cause is that your agent never connects to LiveKit Cloud, so it never publishes an audio track to the room. Below are concise checks and fixes tied to your code.
1) Un-comment and call ctx.connect()
• In entrypoint you commented out:
Copy code
# await ctx.connect(
# # url=LIVEKIT_URL,
# # api_key=LIVEKIT_API_KEY,
# # api_secret=LIVEKIT_API_SECRET,
# ) Copy code
await ctx.connect(
url=LIVEKIT_URL,
api_key=LIVEKIT_API_KEY,
api_secret=LIVEKIT_API_SECRET,
)await session.start(...) Copy code
print("RoomIO created:", session._room_io is not None) # should be True
print("Audio output attached:", session.output.audio is not None) # should be Trueawait session.say("...")tts=DeepgramTTS(...) Copy code
room.on(RoomEvent.ParticipantConnected, p => console.log("Participant:", p.identity));
room.on(RoomEvent.TrackSubscribed, (track, pub, participant) => {
console.log("TrackSubscribed:", { kind: track.kind, source: pub.source, participant: participant.identity });
});autoplayplaysInlinespeech_createdtrack_publishedctx.connect()d
delightful-hamburger-90968
12/20/2025, 10:05 PMwhy did I got this error ?
Traceback (most recent call last):
File "C:\Users\LucNM\AppData\Local\miniconda3\envs\voice-core\lib\site-packages\opentelemetry\util\_decorator.py", line 71, in async_wrapper
return await func(*args, **kwargs) # type: ignore
File "C:\Users\LucNM\AppData\Local\miniconda3\envs\voice-core\lib\site-packages\livekit\agents\ipc\job_proc_lazy_main.py", line 286, in _traceable_entrypoint
await self._job_entrypoint_fnc(job_ctx)
File "C:\tmp\02_Voice\02_Research\01_LiveKit\voice_agent.py", line 43, in entrypoint
await ctx.connect(
TypeError: JobContext.connect() got an unexpected keyword argument 'url'
Here code I used
await ctx.connect(
url=LIVEKIT_URL,
api_key=LIVEKIT_API_KEY,
api_secret=LIVEKIT_API_SECRET,
)
t
tall-airline-16756
12/20/2025, 10:05 PMThanks — I see why the TypeError happened. You're passing connection credentials to
ctx.connect()JobContext.connect()urlapi_keyapi_secretctx.connect() Copy code
await ctx.connect() Copy code
from livekit.agents import AutoSubscribe
await ctx.connect(auto_subscribe=AutoSubscribe.AUDIO_ONLY)ctx.connect()await ctx.connect()auto_subscribe2 Views