fix(voice): cancel realtime generation when speech is interrupted#5703
Open
longcw wants to merge 1 commit into
Open
fix(voice): cancel realtime generation when speech is interrupted#5703longcw wants to merge 1 commit into
longcw wants to merge 1 commit into
Conversation
When session.interrupt() runs before rt_session has emitted response.created, the in-flight generate_reply was forging ahead and the server would still produce a response that the agent then ignored. _realtime_reply_task now races the generate_reply future against the speech interrupt and cancels it on interrupt, and each realtime plugin's generate_reply future now signals a server-side cancel (or drops the queued send) from its done callback.
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| if speech_handle.interrupted: | ||
| # cancel the pending generation; the plugin emits response.cancel | ||
| if not generate_reply_fut.done(): | ||
| generate_reply_fut.cancel() |
Member
There was a problem hiding this comment.
if the future is done, should we call self._rt_session.interrupt() instead?
| self._pending_generation_fut = None | ||
| if f.cancelled() and is_current: | ||
| # external cancel: signal interrupt to Gemini via activity_start | ||
| self.interrupt() |
Member
There was a problem hiding this comment.
This is guarded by _manual_activity_detection in start_user_activity, should we send it here directly?
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
session.interrupt()was a silent no-op when fired before the realtime server had emittedresponse.created. The localSpeechHandlewas marked interrupted, but_realtime_reply_taskhad already passed its auth-wait block and continued to callrt_session.generate_reply(), after which the server produced a response the agent then discarded. Reported in #5642 (comment).Two changes fix it:
_realtime_reply_tasknow races thegenerate_replyfuture againstspeech_handle._interrupt_futviawait_if_not_interrupted. On interrupt, the future is cancelled and the task returns.generate_replyfuture now wires anadd_done_callbackthat, on cancellation, cleans up tracking state and signals the server (or drops the queued send) where the API supports it:response.cancel.interrupt()to sendactivity_startto Gemini.urgency=immediate, defer_response=True).Added
test_generate_reply_cancellationintests/test_realtime/test_realtime.py(parameterized over OpenAI/Azure) which cancels agenerate_replyfuture immediately and verifies a subsequentgenerate_replydoes not hitconversation_already_has_active_response.Fixes #5642