Skip to content

fix: resolve transcript-only amd greetings at eot - #7026

Open
chenghao-mou wants to merge 1 commit into
mainfrom
chenghao/fix/AGT-3382-amd-stt-vad-desync
Open

fix: resolve transcript-only amd greetings at eot#7026
chenghao-mou wants to merge 1 commit into
mainfrom
chenghao/fix/AGT-3382-amd-stt-vad-desync

Conversation

@chenghao-mou

Copy link
Copy Markdown
Member

Problem: AMD can receive final transcripts without VAD boundaries. Its LLM verdict then remains behind the silence gate until the detection timeout.

Behavior: The existing transcript EOT backstop now supplies the missing silence signal only when no valid VAD end exists. None reuses session STT or LLM models, while omitted values keep cloud defaults. Timeout verdicts retain the transcript.

Fixes #6996
Addresses AGT-3382

Initial prompt and agent context

Model: GPT-5.6

Can you investigate #6996?

Yeah, EOT sounds good.

we could have None means session STT or LLM, so NOT_GIVEN -> default, None -> Inherit, explicit -> explicit.

Yeah, we should preserve the transcript and add tests against the regression cases.

but eot requires VAD too, right?

yes please.

@chenghao-mou
chenghao-mou requested a review from a team as a code owner August 28, 2026 12:43

@devin-ai-integration devin-ai-integration Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ Devin Review: No Issues Found

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

Devin Review

Co-authored-by: Jonah Lipsky <36577398+jonahlipsky@users.noreply.github.com>
@chenghao-mou
chenghao-mou force-pushed the chenghao/fix/AGT-3382-amd-stt-vad-desync branch from 8f4d560 to 00fcc7d Compare August 28, 2026 12:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

AMD: quick-greeting calls strand at the detection timeout when the STT and VAD input streams desync

1 participant