hotato GitHubGet started

Endpoint misdetection

acknowledged

10 independent reportsacross 3 sourceson 11 stackslast reported 2026-08-06

What this is

End-of-turn detection fires at the wrong moment, so the agent either answers before the caller has finished or waits after they have.

Status acknowledged: the thread it was reported in was acknowledged upstream.

Evidence

Nasrullah-Ahmed2026-08-06deepgramlivekitsileroGitHub issuesgithub.com/…/6726
Bug Description With turn_detection="stt" and an explicitly configured VAD, min_endpointing_delay does not run after the STT's EndOfTurn signal. It is measured from the earlier moment the speech itself ended,...

First seen 2026-08-06 · our copy fetched 2026-08-06T08:06:38Z · SHA-256 of that copy begins 83863ae585498701… · profile

whitepepper3552026-08-02kokoronemowhisperGitHub issuesgithub.com/…/412
...after that is dropped as input-after-final, even though the user is still speaking and VAD has not detected the end of speech yet. ## Root cause STT/base_stt_handler.py's before_emit_output locks a...

First seen 2026-08-02 · our copy fetched 2026-08-02T12:10:51Z · SHA-256 of that copy begins 52d2cfc6ac1d12ec… · profile

Propconnect-ai2026-07-14cartesiadeepgramlivekitGitHub issuesgithub.com/…/6415
...report — I may well be misreading the design intent. ### What I observed turn-detector-v1, English, unlikely_threshold: 0.56, no overrides. The same word, spoken twice, in two sessions — same...

First seen 2026-07-14 · our copy fetched 2026-07-25T17:50:25Z · SHA-256 of that copy begins e96eff80b1a938ce… · profile

Pulkit07292026-07-12deepgrampipecatGitHub issuesgithub.com/…/5006
...await self.handle_prediction_result(prediction) self.turn_complete = state == EndOfTurnState.COMPLETE # Start the STT timeout (adjusted by VAD stop_secs since that time already elapsed) timeout = max(0, self.stt_timeout - self.stop_secs) The comment...

First seen 2026-07-12 · our copy fetched 2026-07-25T17:50:31Z · SHA-256 of that copy begins 4891230123214ad1… · profile

6 further reports of this mode, same five columns
Marc_Torrellas_Socastro2026-06-10—community.livekit.iocommunity.…ng-vad/1399
I’m getting message: “stt end of speech received while vad is still in a speech segment, flushing vad | extras: vad_speech_start_time=1781089362.6959133, flushed=True” as a warning but it seems like a...

First seen 2026-06-10 · our copy fetched 2026-07-25T18:25:16Z · SHA-256 of that copy begins 29133c36b1e069f5… · profile

TimeHorizon21002026-06-03twilioGitHub issuesgithub.com/…/927
...was filtering silence. - Now forwarding full audio stream including silence to allow model-side VAD/end-of-turn. - Reduced and simplified prompt for faster speech output. - Fixed language typo in prompt...

First seen 2026-06-03 · our copy fetched 2026-07-24T21:00:36Z · SHA-256 of that copy begins fededd97052cf300… · profile

evara-ai2026-03-03deepgramelevenlabstwilioHacker Newsnews.ycomb…id=47230233
...jumping in mid-sentence or waiting too long. Switching from fixed silence thresholds to semantic end-of-turn detection was night and day. One dimension I'd add: geography matters even more when your...

First seen 2026-03-03 · our copy fetched 2026-07-25T17:50:19Z · SHA-256 of that copy begins 7b3c299ade3943aa… · profile

nowittyusername2026-02-16—Hacker Newsnews.ycomb…id=47029721
...him PTT by default as the problems of the model accurately guessing the end of utterance are just too great. I think it can be solved in the future but,...

First seen 2026-02-16 · our copy fetched 2026-07-24T20:53:10Z · SHA-256 of that copy begins 2c0a1d37a6c52fa7… · profile

IngLP2025-11-25elevenlabslivekitsileroGitHub issuesgithub.com/…/4087
...talking, I tried in a room with absolute silence too. ### Expected Behavior Silero VAD commits the turn after detecting user silence. Agent respond to the user message already timely...

First seen 2025-11-25 · our copy fetched 2026-07-30T14:25:59Z · SHA-256 of that copy begins 37e093a57a47ffa5… · profile

Jeeva-MV2025-04-05azure-speechlivekitsileroGitHub issuesgithub.com/…/1893
...any suggestions on how to tackle this will be helpful. Attaching the STT and VAD configs along with a sample log file for reference. Azure-configs (Tried with multi-lingual detection as...

First seen 2025-04-05 · our copy fetched 2026-07-26T01:11:30Z · SHA-256 of that copy begins 1bbeb9bd6df51b2d… · profile

Quotes run to 30 words around the reported failure and link to the post.Add your call to this entryTake a quote down within 24 hours.

What is measurable

SignalThis record stands on its own linked sources. To put a number on your own call, hotato autopsy ./call.wav reads a recording you already have and ranks the timing incidents in it.
EstablishesEvery report below links to the original post and names the handle that wrote it; the quote is at most 30 words around the reported failure.
What else it establishes, and what your own trace settles
  • The status changes only when a re-check stores the evidence link, the quote and the date behind the change.
  • Every timestamp comes from the fetched artifact it describes, recorded with the sha256 of the response body it was read from.

What your trace settles

  • Reproducing the mode on your own stack is the next step: the reports below carry the configuration each reporter was running.
TierAtlas record: every field on this page traces to a fetched public artifact. hotato is MIT and runs locally, getting started.

Status history

DateStateWhat moved it
2026-07-26openacknowledgedissue closed (not_planned) evidence dated 2026-07-26

Discussion

Reader discussion of this mode: what fixed it and what did not. The reports above are attributed and linked; a comment is a comment.

Open this discussion on GitHub

The record

Ids, digests, every link this entry stands on, and how to take a quote down
Atlas idVFA-2026-0002
Mode idpat_4be4cb5af54e4a4aaa9ff1a4
Content digestsha256:4be4cb5af54e4a4aaa9ff1a4d0a837211d90735f9dea2da06e96a9bbe6b4d80a
First reported2025-04-05
Last reported2026-08-06
First seen by the atlas2026-07-21T08:56:59Z
Last re-checked2026-08-06T07:35:56Z
Reported onazure-speechcartesiadeepgramelevenlabskokorolivekitnemopipecatsilerotwiliowhisper
SourcesGitHub issues, Hacker News, community.livekit.io
Scope10 independent public reports across 3 sources. This entry states where the mode has been reported; where else it can occur is what the next report settles.
References
  1. reporthttps://github.com/livekit/agents/issues/6726 · Nasrullah-Ahmed · GitHub issues · 2026-08-06
  2. reporthttps://github.com/huggingface/speech-to-speech/issues/412 · whitepepper355 · GitHub issues · 2026-08-02
  3. reporthttps://github.com/livekit/agents/issues/6415 · Propconnect-ai · GitHub issues · 2026-07-14
  4. reporthttps://github.com/pipecat-ai/pipecat/issues/5006 · Pulkit0729 · GitHub issues · 2026-07-12
  5. reporthttps://community.livekit.io/t/warning-stt-end-of-speech-received-while-vad-is-still-in-a-speech-segment-flushing-vad/1399 · Marc_Torrellas_Socastro · community.livekit.io · 2026-06-10
  6. reporthttps://github.com/twilio/twilio-python/issues/927 · TimeHorizon2100 · GitHub issues · 2026-06-03
  7. reporthttps://news.ycombinator.com/item?id=47230233 · evara-ai · Hacker News · 2026-03-03
  8. reporthttps://news.ycombinator.com/item?id=47029721 · nowittyusername · Hacker News · 2026-02-16
  9. reporthttps://github.com/livekit/agents/issues/4087 · IngLP · GitHub issues · 2025-11-25
  10. reporthttps://github.com/livekit/agents/issues/1893 · Jeeva-MV · GitHub issues · 2025-04-05
Machine surfacesfeed.jsonfeed.atomfeed.schema.jsonatlas.llms.txt
RemovalIf you wrote something quoted on this page and want it gone, one email to [email protected] does it. The removal path takes the quote, the handle and the link off this page, out of the feeds and out of the llms.txt within 24 hours, and tombstones the id so the entry cannot quietly return at a new address.