ERROR asyncio Task exception was never retrieved future lt T LiveKit Community #ask-ai

ERROR asyncio - Task exception was never retrieved...

brief-vase-33757

08/29/2025, 3:56 AM

ERROR asyncio - Task exception was never retrieved future: <Task finished name='Task-431' coro=<AgentSetup.say_greetings() done, defined at /Users/sandeep/projects/pulse/agent/base_agent.py:122> exception=RuntimeError('trying to generate speech from text without a TTS model')>

tall-airline-16756

08/29/2025, 3:56 AM

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

tall-airline-16756

08/29/2025, 3:57 AM

Hey there! The error you're seeing happens because the system is trying to generate speech from text, but it doesn’t have a Text-to-Speech (TTS) model set up yet. To fix this, just make sure you include a TTS model when you initialize the

AgentSession

. Here’s a quick example:

Copy code

session = AgentSession(
    stt=deepgram.STT(model="nova-3", language="multi"),
    llm=openai.LLM(model="gpt-4o-mini"),
    tts=cartesia.TTS(model="sonic-2", voice="f786b574-daa5-4673-aa0c-cbe3e8534c02"),  # This part is important!
    vad=silero.VAD.load(),
    turn_detection=MultilingualModel(),
)

Just double-check that the TTS model you pick works well with your setup. Some popular options are Cartesia, Deepgram, ElevenLabs, Google Gemini, and a few others. Each one needs specific details like

model

and

voice

brief-vase-33757

08/29/2025, 3:58 AM

but I am using session = AgentSession( vad=ctx.proc.userdata["vad"], _min_interruption_words_=0, userdata=UserData(), _min_endpointing_delay_=0.08, _max_endpointing_delay_=0.80, _max_tool_steps_=10, _user_away_timeout_=SILENCE_THRESHOLD, llm=realtime.RealtimeModel() # Using realtime model here )

tall-airline-16756

08/29/2025, 3:59 AM

When you’re working with a realtime model in

AgentSession

, it’s important to set it up for text-only output and add a separate TTS plugin for turning that text into speech. Here’s a friendly little guide to get you started:

Copy code

from livekit.agents import AgentSession
from livekit.plugins import openai

session = AgentSession(
    llm=openai.realtime.RealtimeModel(modalities=["text"]),  # Text-only output
    tts=openai.TTS(voice="ash")  # Add a TTS plugin
)

This way, your realtime model focuses on generating text responses, and the TTS plugin takes care of converting those into speech. You’ve got options like OpenAI, Cartesia, Deepgram, and more for TTS. If you want a hand picking the right TTS provider or tweaking any settings, just give me a shout—I’m here to help! Sources: Text-to-speech (TTS) integrations | LiveKit Docs | OpenAI Realtime API integration guide | LiveKit Docs | Turn detection and interruptions | LiveKit Docs | OpenAI TTS integration guide | LiveKit Docs | Realtime model integrations | LiveKit Docs

Open in Slack

Previous Next