why am I not hearing my agent talk whenever calling a genera LiveKit Community #ask-ai

Join Slack

why am I not hearing my agent talk whenever callin...

# ask-ai

wide-dress-57902

08/14/2025, 3:34 AM

why am I not hearing my agent talk whenever calling a generate reply in the session?

tall-airline-16756

08/14/2025, 3:34 AM

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

tall-airline-16756

08/14/2025, 3:34 AM

The issue where the agent's voice isn’t coming through during a session when calling

generate_reply()

can happen for a few reasons: • TTS Configuration: Make sure the Text-to-Speech (TTS) plugin is set up correctly. Without it, the agent won’t be able to produce any audio. • Realtime Model Usage: If you’re working with a realtime model, double-check that you’ve added a TTS plugin to the session. Realtime models usually need this to actually speak. • Audio Enabled Settings: Take a quick look at your

RoomOutputOptions

and confirm that

audio_enabled=True

is set. Here’s how you can fix it: 1. Make sure both the TTS plugin and the realtime model (if you’re using one) are properly set up in your

AgentSession

. 2. If

generate_reply()

isn’t giving you audio, try using

session.say()

for your pre-defined messages instead. 3. Double-check that audio output is turned on in your settings. If you want, I’d be happy to share an example configuration to help you get the TTS and realtime model working smoothly. Just let me know! Sources: Agent speech and audio | LiveKit Docs | tests/test_agent_session.py | Text and transcriptions | LiveKit Docs

wide-dress-57902

08/14/2025, 3:53 AM

say method is working but not generate_reply method, so not sure why that is

tall-airline-16756

08/14/2025, 3:54 AM

Thanks for letting me know that

session.say()

is working! To get to the bottom of why

generate_reply()

isn’t producing audio, could you share a bit more about your setup? • Are you using a realtime model with

generate_reply()

? • Do you have a TTS (Text-to-Speech) plugin set up for the session? • Have you turned on audio output in

RoomOutputOptions

? With these details, I’ll be able to help figure out what’s going on and find a fix for you.

wide-dress-57902

08/14/2025, 3:55 AM

I am not using realtime model. I do have a TTS setup which is working for everything else. I have not configured anything with RoomOutputOptions

tall-airline-16756

08/14/2025, 3:55 AM

Since you’re not using a realtime model and your TTS setup works with other methods, the issue probably isn’t related to

RoomOutputOptions

. These options are usually set up at the session level during initialization and don’t need to be included with every

generate_reply()

call. To help figure this out, here are a few things you might want to check: • Make sure the TTS plugin is compatible with

generate_reply()

. Sometimes, a TTS setup works fine with

session.say()

but not with

generate_reply()

. • Double-check that the input text you’re passing to

generate_reply()

is valid and doesn’t have any unsupported characters or formats. • Take a look at the logs for any errors or warnings when you call

generate_reply()

. They often give helpful hints about what might be going wrong. If you’d like, I’m here to help review your TTS configuration or go through the logs with you. Just let me know! Sources: examples/voice_agents/realtime_load_chat_history.py | Agent speech and audio | LiveKit Docs

Open in Slack

Previous Next