When calling

session.say()

inside

tts_node()

, what guarantees exist about suppression of further yields? Is there an integrated mechanism to prevent

tts_node()

from continuing and yielding after a

session.say()

call?

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

You're asking how LiveKit reconciles

tts_node()

generator yields with

session.say()

invoked in the same agent turn, and how to avoid double/overlapping speech. Short answer: LiveKit does not automatically reconcile them. If

session.say()

is called while

tts_node()

is yielding, both can produce audio unless you coordinate them. Recommended patterns to avoid double speech • Use one mechanism per turn: ◦ Prefer

tts_node()

for LLM-driven replies. ◦ Use

session.say()

only for injected/short phrases (greetings, interrupts). • Return or stop the generator after calling `session.say()`: ◦ Call

return

from

tts_node()

or set a control flag checked before yielding further frames. • Interrupt/cancel current speech before calling `session.say()`: ◦ Call

session.interrupt()

or interrupt the active

SpeechHandle

, then

await

cancellation if needed, then call

session.say()

. • Chain via separate turns when injecting speech: ◦ Interrupt →

session.say()

→ start a new

generate_reply()

if follow-up LLM output is needed. • Use pre-synthesized audio to bypass TTS: ◦ Call

session.say(audio=...)

with prepared frames so

tts_node()

is not involved. If you tell me which behavior you want (immediate stop, finish current chunk then stop, or cancel queued speech), I’ll provide a concise code pattern you can plug into your

tts_node()

implementation. Sources: Agent speech and audio | LiveKit docs | Pipeline nodes and hooks | LiveKit docs | Interrupt User | LiveKit docs | Turns overview | LiveKit docs | Text and transcriptions | LiveKit docs