From 9d8eefd2a231b09dfbea4456f1c00c1d0fa57f4d Mon Sep 17 00:00:00 2001 From: Mark Backman Date: Mon, 20 Apr 2026 11:58:20 -0400 Subject: [PATCH] Add changelog for #4337 --- changelog/4337.changed.2.md | 1 + changelog/4337.changed.md | 1 + 2 files changed, 2 insertions(+) create mode 100644 changelog/4337.changed.2.md create mode 100644 changelog/4337.changed.md diff --git a/changelog/4337.changed.2.md b/changelog/4337.changed.2.md new file mode 100644 index 000000000..a1da061c8 --- /dev/null +++ b/changelog/4337.changed.2.md @@ -0,0 +1 @@ +- `SpeechTimeoutUserTurnStopStrategy` now waits only `user_speech_timeout` when a transcript arrives without a VAD stop event, rather than `max(ttfs_p99_latency, user_speech_timeout)`. If you had `ttfs_p99_latency > user_speech_timeout`, turn detection in that path is slightly faster than before. diff --git a/changelog/4337.changed.md b/changelog/4337.changed.md new file mode 100644 index 000000000..67c1ddbfe --- /dev/null +++ b/changelog/4337.changed.md @@ -0,0 +1 @@ +- If you use an STT service that emits finalized transcripts (Speechmatics, Soniox, Deepgram Flux, AssemblyAI) with `SpeechTimeoutUserTurnStopStrategy`, user turns now end as soon as `user_speech_timeout` elapses after VAD stop. Previously the strategy also waited for the STT P99 latency (`ttfs_p99_latency`) even when the transcript was already marked final. `user_speech_timeout` is still honored as a floor — STT finalization never shortens it.