Feat/voice robustness - #99

Merged

berges99 merged 15 commits into

mainfrom

feat/voice-robustness

Jul 24, 2026

Commits on Jul 23, 2026

Commits on Jul 24, 2026

fix(voice): Namo text EOU, shorter complete-text holds, fix hold grace stretch
berges99
committed
test(voice): move voice/turn tests under python/tests/voice
berges99
committed
fix(voice): ignore STT crumbs after finished commits, hide vetoed partials
berges99
committed
feat(voice): Deepgram STT, playground model/STT controls, fix long-reply barge-in
berges99
committed
fix(voice): align LLM warmup/STT fallback; don't drop mid-turn CONTINUE
berges99
committed
fix(voice): send config STT provider id on session_started; harden HOLD defer test
berges99
committed
fix(voice): serialize Deepgram STT websocket sends under one wire lock
berges99
committed
test(voice): assert HOLD deferral on processed partials, not inject wall-clock
berges99
committed
fix(voice): don't treat contentful "...you know/I mean" as EOU hedges
berges99
committed
test(voice): close HOLD-defer race before first processed partial
berges99
committed
test(voice): stop gating HOLD deferral on _last_commit_at
berges99
committed
test(voice): wait for partial text, not monotonic tick advance
berges99
committed