Which ways can I check in my agent if speaking out the respo LiveKit Community #ask-ai

Which ways can I check in my agent: - if speaking ...

chilly-spring-61928

12/15/2025, 7:52 PM

Which ways can I check in my agent: • if speaking out the response has already been started • if the current response has been interrupted

tall-airline-16756

12/15/2025, 7:52 PM

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

tall-airline-16756

12/15/2025, 7:53 PM

You're asking how to detect (1) when the agent has started speaking and (2) whether the current response was interrupted. Two reliable ways: • Use the SpeechHandle returned by session.say() or session.generate_reply() ◦ The returned handle indicates speech generation/playout has begun. ◦ Check

handle.interrupted

after playback (or when the handle resolves) to know if it was interrupted. ◦ Examples: ▪︎ Python:

Copy code

handle = session.say("Hello")
      await handle  # waits for playback to complete
      if handle.interrupted:
          print("Speech was interrupted")

• TypeScript:

Copy code

const handle = session.say("Hello");
      await handle; // waits for completion
      if (handle.interrupted) console.log("Speech was interrupted");

• Listen for agent state changes (agent_state_changed) ◦ Monitor the event and check for

new_state == "speaking"

to detect when speaking starts. ◦ Examples: ▪︎ Python:

Copy code

@session.on("agent_state_changed")
      def on_agent_state_changed(ev):
          if ev.new_state == "speaking":
              print("Agent started speaking")

• TypeScript:

Copy code

session.on(voice.AgentSessionEventTypes.AgentStateChanged, (ev) => {
        if (ev.newState === "speaking") console.log("Agent started speaking");
      });

Notes: • Interruptions occur when the user starts speaking and

allow_interruptions=True

(default), which stops playback and marks the SpeechHandle as interrupted. • If you want, tell me which SDK/language you’re using and whether you prefer event-driven checks or polling, and I’ll provide a short copy-paste snippet for that environment. Sources: Agent speech and audio | LiveKit docs | Turn detection and interruptions | LiveKit Docs

chilly-spring-61928

12/15/2025, 7:57 PM

So how could I check in the agents transcription_node of the response that the chunks belong to: • is already being spoken • Is interrupted

tall-airline-16756

12/15/2025, 7:58 PM

You're asking whether, inside the agent's

transcription_node

, you can tell if a given transcription chunk is (a) already being spoken or (b) was interrupted. Short answer: you cannot reliably get real-time speaking or interruption status from inside

transcription_node

. It only receives transcription text (or

TimedString

when TTS-aligned transcripts are enabled) for processing; it does not expose the runtime playback state or the

SpeechHandle

interruption flag. What you can do instead (recommended): • Use

TimedString

with

use_tts_aligned_transcript=True

to get word-level `start_time`/`end_time` metadata for when words are scheduled to be spoken. This tells you when words are planned to play, not whether playback actually completed. • Track the actual runtime state and interruptions outside the `transcription_node`: ◦ Monitor

agent_state_changed

events to detect when the agent enters

speaking

. ◦ Track the

SpeechHandle

returned by

session.say()

session.generate_reply()

and check

handle.interrupted

after playback or when the handle resolves. ◦ Use the

speech_created

event to get and store the

SpeechHandle

for the current utterance so you can correlate it with transcription output. If helpful, tell me: • Which SDK/language you’re using (Python/TypeScript), and • Whether you want examples showing how to correlate

TimedString

timestamps with a

SpeechHandle

and

agent_state_changed

2 Views

Open in Slack

Previous Next