fix(voice): clean up turn variables when committing a user turn manually

[toubatbrian]

Summary

Automated port of livekit/agents#5671 into agents-js.

When clearUserTurn() is invoked (the path used by manual turn detection to abandon an in-progress turn), it previously left several turn-scoped fields stale:

• lastFinalTranscriptTime, lastSpeakingTime, speechStartTime would carry over into the next turn, corrupting the EOU transcriptionDelay / endOfUtteranceDelay metrics and the preemptive-generation startedSpeakingAt for the next turn.
• The current user_turn span was kept open and would be reused for the next user turn instead of starting a fresh one. Any provider STT request ids collected so far would also be flushed onto the wrong span.

This fix resets those turn variables and ends the in-progress user_turn span so the next speech starts a fresh span and metrics window.

cc @toubatbrian @livekit/agent-devs

Ported features

• agents/src/voice/audio_recognition.ts — extends clearUserTurn() to:
  • reset lastFinalTranscriptTime to 0, speechStartTime and lastSpeakingTime to undefined (mirrors _last_final_transcript_time = None, _speech_start_time = None, _last_speaking_time = None in Python)
  • end the in-progress userTurnSpan (if recording) and clear the cached span + accumulated sttRequestIds, so the next speech starts a fresh span and the next ended span doesn't inherit stale provider request ids

Implementation nuances vs. Python

• _vad_speech_started field has no JS equivalent. In Python, _vad_speech_started is a separate boolean used to gate _speech_start_time assignment in the VAD START_OF_SPEECH handler. The Python diff resets both fields together (self._speech_start_time = None, self._vad_speech_started = False).

  In agents-js, that gate is implemented directly as if (this.speechStartTime === undefined) in the VAD INFERENCE_DONE handler — there is no separate vadSpeechStarted field. Resetting speechStartTime = undefined therefore re-arms the same gate, so the Python _vad_speech_started = False line maps to a no-op in JS. No new field was introduced.

• lastFinalTranscriptTime sentinel value. Python uses float | None; agents-js types this as number and uses 0 as the "not set" sentinel (already established in bounceEOUTask's lastFinalTranscriptTime !== 0 check and in the existing endUserTurn reset path that assigns this.lastFinalTranscriptTime = 0). The port resets to 0 for consistency rather than introducing undefined.

• update_stt(None) / update_stt(stt) reset. The existing restartStt() IIFE already mirrors the Python update_stt reset (it stops the STT tasks, closes the pipeline, then re-starts the pipeline), so no change was needed there. It's preserved as-is and runs after the new resets.

Test plan

• pnpm build:agents succeeds.
• pnpm exec eslint agents/src/voice/audio_recognition.ts clean.
• pnpm exec prettier --check clean for the touched file and changeset.
• pnpm exec vitest run agents/src/voice/audio_recognition_span.test.ts (4 passed).
• Manual verification with manual turn detection: trigger clearUserTurn mid-turn, confirm the next turn produces a fresh user_turn span and correct EOU metrics.

https://claude.ai/code/session_01P6kkyq5V2krvxdChbADKzg

---

Generated by Claude Code

[toubatbrian]

/cc @toubatbrian @livekit/agent-devs for review.

---

Generated by Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: e4aa7ce

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

This PR includes changesets to release 31 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-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

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[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 3 additional findings.