preemptive generation feature #783
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
7 tasks
toubatbrian
reviewed
Oct 24, 2025
Comment on lines
+515
to
+534
| isEquivalent(other: ChatContext): boolean { | ||
| // Same object reference | ||
| if (this === other) { | ||
| return true; | ||
| } | ||
|
|
||
| // Different lengths | ||
| if (this._items.length !== other._items.length) { | ||
| return false; | ||
| } | ||
|
|
||
| // Compare each item pair | ||
| for (let i = 0; i < this._items.length; i++) { | ||
| const a = this._items[i]!; | ||
| const b = other._items[i]!; | ||
|
|
||
| // IDs and types must match | ||
| if (a.id !== b.id || a.type !== b.type) { | ||
| return false; | ||
| } |
There was a problem hiding this comment.
This is nice, can you also add a unittest for this function? Inside chat_context.test.ts?
toubatbrian
reviewed
Oct 24, 2025
Comment on lines
+1054
to
+1087
| const preemptive = this.preemptiveGeneration; | ||
| if (preemptive) { | ||
| // Add the user message to the chat context for comparison | ||
| const validationChatCtx = this.agent.chatCtx.copy(); | ||
| if (userMessage) { | ||
| validationChatCtx.insert(userMessage); | ||
| } | ||
|
|
||
| // Validate: transcript matches, context equivalent, tools unchanged, toolChoice unchanged | ||
| const transcriptMatches = preemptive.info.newTranscript === info.newTranscript; | ||
| const contextEquivalent = preemptive.chatCtx.isEquivalent(validationChatCtx); | ||
| const toolsUnchanged = preemptive.tools === this.agent.toolCtx; | ||
| const toolChoiceUnchanged = preemptive.toolChoice === this.toolChoice; | ||
|
|
||
| if (transcriptMatches && contextEquivalent && toolsUnchanged && toolChoiceUnchanged) { | ||
| // Use preemptive generation! | ||
| const speechHandle = preemptive.speechHandle; | ||
| this.preemptiveGeneration = undefined; | ||
|
|
||
| const leadTime = Date.now() - preemptive.createdAt; | ||
| this.logger.info( | ||
| { | ||
| transcript: info.newTranscript, | ||
| leadTimeMs: leadTime, | ||
| confidence: preemptive.info.transcriptConfidence, | ||
| }, | ||
| 'using preemptive generation', | ||
| ); | ||
|
|
||
| // Schedule the preemptive speech | ||
| this.scheduleSpeech(speechHandle, SpeechHandle.SPEECH_PRIORITY_NORMAL); | ||
|
|
||
| // Emit metrics | ||
| const eouMetrics: EOUMetrics = { |
There was a problem hiding this comment.
Can we make sure we have the parity implementation as in python agent framework? https://github.com/livekit/agents/blob/a9bc03562f498f3666978ad008fc93b2cbbd22a9/livekit-agents/livekit/agents/voice/agent_activity.py#L1384-L1420
Comment on lines
+236
to
+245
| // Update preflight transcript and confidence | ||
| this.audioPreflightTranscript = `${this.audioTranscript} ${preflightTranscript}`.trim(); | ||
| this.preflightTranscriptConfidence = preflightConfidence; | ||
|
|
||
| // Trigger preemptive generation if conditions are met | ||
| if ( | ||
| this.hooks.onPreemptiveGeneration && | ||
| (this.turnDetectionMode !== 'manual' || this.userTurnCommitted) | ||
| ) { | ||
| // Calculate confidence including all final transcripts plus the current preflight |
There was a problem hiding this comment.
Let's follow the same params naming as in python agent:
# still need to increment it as it's used for turn detection,
self._last_final_transcript_time = time.time()
# preflight transcript includes all pre-committed transcripts (including final transcript from the previous STT run)
self._audio_preflight_transcript = (self._audio_transcript + " " + transcript).lstrip()
self._audio_interim_transcript = transcript
if not self._vad or self._last_speaking_time == 0:
# vad disabled, use stt timestamp
self._last_speaking_time = time.time()
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
this allows STTs to send PREFLIGHT_TRANSCRIPT events.
"exact" same implemenation as the python library.
see #773
Changes Made
brings the SpeechEventType PREFLIGHT_TRANSCRIPT to the typescript library. but besides this it also enables preempetive generation before VAD end-of-speech or turn detection completes, to start generation early.
Btw: I couldn't find any actual PREFLIGHT_TRANSCRIPT events in the python version... Am I missing something?I found it ;-) Deepgram STT feature support for preemptive gen is prepared here: simllll/agents-js@feat/preemtive-gen...simllll:agents-js:feat/preemptive-gen-deepgram-stt
Pre-Review Checklist
Testing
restaurant_agent.tsandrealtime_agent.tswork properly (for major changes)Additional Notes
Note to reviewers: Please ensure the pre-review checklist is completed before starting your review.