Fix playback flush and speech interruption races

[toubatbrian]

Summary

Fixes an STT-deafness wedge that occurred when a handoff cascade fires three
_updateActivity calls in <10 ms (e.g. TaskGroup resolving while the next task
starts to greet). In that window, the caller's speech triggered
interruptByAudioActivity() during the thinking → speaking transition, and a
defensive wasAgentSpeaking guard skipped clearing isAgentSpeaking — leaving
AudioRecognition permanently holding every subsequent STT event in
transcriptBuffer. The agent appeared dead for ~90 s until the false-interrupt
timer eventually unwedged it.

The fix aligns the JS implementation with the Python upstream
(livekit-agents/livekit/agents/voice/agent_activity.py) for the affected paths
and adds two narrow JS-only defenses.

agent_activity.ts — align with Python

• interruptByAudioActivity() now always calls onEndOfAgentSpeech() and
  _updateAgentState('listening') whenever the pause path runs, instead of
  guarding on agentState === 'speaking'. This was the root cause of the wedge:
  during the thinking → speaking transition the guard was false and the
  STT-hold flag was never cleared. Python's
  _interrupt_by_audio_activity does this unconditionally.
• onPipelineReplyDone() now also calls onEndOfAgentSpeech() and
  restoreInterruptionByAudioActivity() (matches Python's
  _on_pipeline_reply_done) as a safety net at the end of every reply task.
• Tool-reply continuation now early-returns and cancels executeToolsTask when
  the speech handle is interrupted, matching Python's
  if speech_handle.interrupted: await cancel_and_wait(exe_task); return.
• The 'thinking' transition is gated on !speechHandle.interrupted, matching
  Python's if not speech_handle.interrupted and len(tool_output.output) > 0.

audio_recognition.ts — JS async-only fixes

• onEndOfAgentSpeech() clears isAgentSpeaking = false before awaiting
  the interruption sentinel write. Python is unaffected because it uses
  send_nowait; in JS the awaitable interruptionStreamChannel.write can stall
  and let new STT events get buffered against an agent that has already
  finished speaking.
• onEndOfOverlapSpeech() now uses a snapshot of ignoreUserTranscriptUntil
  captured before the in-function mutation. The previous condition compared the
  already-mutated value and so was unreachable, suppressing the overlap-end
  notification for the rest

[changeset-bot]

🦋 Changeset detected

Latest commit: 84c7298

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 32 packages

Name	Type
@livekit/agents	Patch
@livekit/agents-plugin-anam	Patch
@livekit/agents-plugin-assemblyai	Patch
@livekit/agents-plugin-baseten	Patch
@livekit/agents-plugin-bey	Patch
@livekit/agents-plugin-cartesia	Patch
@livekit/agents-plugin-cerebras	Patch
@livekit/agents-plugin-deepgram	Patch
@livekit/agents-plugin-elevenlabs	Patch
@livekit/agents-plugin-fishaudio	Patch
@livekit/agents-plugin-google	Patch
@livekit/agents-plugin-hedra	Patch
@livekit/agents-plugin-hume	Patch
@livekit/agents-plugin-inworld	Patch
@livekit/agents-plugin-lemonslice	Patch
@livekit/agents-plugin-liveavatar	Patch
@livekit/agents-plugin-livekit	Patch
@livekit/agents-plugin-minimax	Patch
@livekit/agents-plugin-mistral	Patch
@livekit/agents-plugin-mistralai	Patch
@livekit/agents-plugin-neuphonic	Patch
@livekit/agents-plugin-openai	Patch
@livekit/agents-plugin-phonic	Patch
@livekit/agents-plugin-resemble	Patch
@livekit/agents-plugin-rime	Patch
@livekit/agents-plugin-runway	Patch
@livekit/agents-plugin-sarvam	Patch
@livekit/agents-plugin-silero	Patch
@livekit/agents-plugin-tavus	Patch
@livekit/agents-plugins-test	Patch
@livekit/agents-plugin-trugen	Patch
@livekit/agents-plugin-xai	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 5 additional findings.