release: v2.2.3 — hf_xet, CPU-bloat filter, loopback-lost warning

release: v2.2.2 — voiceprint transfer between machines

release: v2.2.1 stable — meeting quality/speed + module bundling fix

release: v2.3.0 stable — meeting quality/speed + module bundling fix

perf(diar): batch WeSpeaker embeddings in Stage O

perf(file): parallelize diarisation and full-audio ASR pass

fix(meeting): convert multilingual language_mode to None for Whisper full-audio pass

release: v2.3.0-beta1 — meeting quality/speed pass

release: v2.2.0 stable

test: build-time selfcheck + frozen-exe gate + unit suite

build: BUILD_SHA = 9f04ec1 for v2.1.0-beta16

fix(cuda,frozen): preload nvidia DLLs from _MEIPASS too

build: BUILD_SHA = 7ecb2a5 for v2.1.0-beta15

build: BUILD_SHA = 58dfa49 for v2.1.0-beta5

docs: TECH_DEBT — Stage AI fine-tune caps/punct + retrain plan

feat(file+ptt+meeting): full-audio file ASR, sequential queue, solo-collapse

chore(assets): regenerate icons from new logo (multi-size .ico/.icns/.iconset)

build: BUILD_SHA = 60ed9c3 for v2.1.0-beta4

feat(meeting,v6): ONNX MossFormer + selective separation + queued finalize + 7-tab settings

feat(tray): "Записи" submenu — last 5 meetings + open folder

build: BUILD_SHA = 5edc557 for v2.1.0-beta3

fix(frozen): torch stub now overrides PyInstaller empty placeholder