how much lowest latency can we get using livekit ? minimum ?

Thanks for reaching out, Av. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

wow superb . how ro archive that level latency ?

To help you get sub-100ms latency with LiveKit, here are some friendly tips: • Use LiveKit Cloud: It’s got a global edge network and multi-home setup that smartly routes your data to cut down delays. • Optimize Network Connections: For control devices, wired Ethernet connections work best, and for mobile units, using the latest WiFi standards (WiFi 6/7) can really help keep local network delays low. • Deploy Close to Users: Make sure your users connect to the edge server that’s geographically nearest to them — this can shave off some valuable milliseconds. • Minimize Public Internet Usage: Whenever you can, route media through private backbone connections to dodge the usual public internet slowdowns. • Streamline Processing: Try to cut down processing delays at every step by tuning your hardware and software setups. If you want a hand putting these into practice or have any specific questions about your deployment, just give me a shout—I’m here to help!

what TTS,STT,LLM you suggest archive that level latency

If you're aiming for low latency in TTS (Text-to-Speech), STT (Speech-to-Text), and LLM (Large Language Model) inference, here are some friendly recommendations based on LiveKit's performance benchmarks: • STT: Try using whisper-large-v3-turbo, which showed an STT latency of about 800ms. • LLM: Go for llama-3.3-70b-versatile, which reached a time-to-first-token (TTFT) latency around 200ms. • TTS: Consider playai-tts, with a TTS latency near 350ms. These components were tested in optimized pipelines and work really well with LiveKit's infrastructure. If you have any specific needs or want a hand setting these up, just give me a shout—I'm here to help!

load_dotenv(override=True) instructions = ( "You are a helpful voice AI assistant." ) class Assistant(Agent): def __init__(self) -> None: super().__init__(instructions=instructions) async def entrypoint(ctx: agents.JobContext): session = AgentSession( llm=openai.realtime.RealtimeModel(model="gpt-4o-mini-realtime-preview",modalities=["text"]), tts=openai.TTS( model="tts-1", voice="ash", ), ) await session.start( room=ctx.room, agent=Assistant(), room_input_options=RoomInputOptions( # For telephony applications, use

BVCTelephony

instead for best results noise_cancellation=noise_cancellation.BVC(), ), ) if name == "__main__": agents.cli.run_app(agents.WorkerOptions(entrypoint_fnc=entrypoint,port=1100)) this is my current code. i run only cli version. how archive that level latency

I don't have the answer you're looking for. You could also try asking your question: • in one of the other Slack channels or • to https://deepwiki.com/livekit/livekit_composite which is trained on all LiveKit source code If you find the answer, please post it here to help others!

ok got it . what STT,LLM,TTS services you recomend for me archive less than 1 second latency ?