Only the TTS model (~2.4 GB) is required on first run; ASR models are
per-platform curated picks (curated_on in models.yaml) installed on demand.
Every transcription surface returns a typed asr_model_missing error with a
one-click download CTA instead of silently pulling multi-GB Whisper weights.
Settings -> Models is a grouped, platform-aware catalog. New guided
permissions UX (wizard System Check + Settings -> Permissions + mic
pre-flight) with native mic-state checks and OS settings deep-links. New
parakeet-mlx engine brings Parakeet TDT v3 to Apple Silicon (language-gated
capture preference so multilingual dictation never regresses). Docs:
expressive-speech page, Flush/Unload + CPU-fallback triage, clone-length FAQ.
Hardening: preflight fails open for custom model pins, ROCm curation no
longer inherits NVIDIA picks, Windows mic probe reads the NonPackaged
consent key, CaptureWidget setup race fixed, offline-cache CI simulation
fixes so empty-cache runners stay green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
P0 — Refinement blocked every dictation final with no timeout. With refinement
auto:true and a slow/dead LLM endpoint, maybe_refine ran unbounded and blocked
the final send in all three capture_ws handlers (~51s measured; the pill hung
"Transcribing…" until the widget's 15s fallback fired). Fix the class: a hard,
env-tunable budget (OMNIVOICE_REFINE_TIMEOUT_S, default 4s) via a new
maybe_refine_async — a slow/dead endpoint now falls back to the unrefined (but
polished) text within the budget and can NEVER delay the final beyond it. The
LLM HTTP call is bounded to the same budget so the orphaned worker unwinds
instead of holding a connection for the client's full 45s. Refinement is now
also fully best-effort in the legacy handler (it can't turn a good final into
an error frame).
P1 — REST /transcribe lacked polish parity. capture.py never applied
polish_text, so REST returned raw "…test" while the WS returned "…test."
Apply text_polish.polish_text to `text` and `refined_text` (segments stay raw),
so the widget POST fallback and MCP/CLI callers match the live socket.
P1 — The #888 "instant first dictation" preload was a no-op. The preload called
warmup() only `if hasattr`, but SherpaDictationBackend had none, and the WS
handlers built a FRESH backend per session so a warm singleton wasn't reused.
Add SherpaDictationBackend.warmup() (builds the recognizer) and share one warm
recognizer per model id across sessions (get_sherpa_dictation_backend, same
invalidation + a shared lock as the capture singleton); each session keeps its
own decode stream. First dictation no longer pays the 1.3–2.5s load.
P1 — llm_ready is a lie (feeds the P0). It only means "an endpoint is
configured", so a placeholder key reads as ready. The P0 timeout makes a dead
endpoint harmless; add last_refine_status so RefinementPanel flags a
configured-but-failing LLM and links to LLM Providers → Test.
Regression tests (fail-before/pass-after): slow-LLM WS final arrives < budget;
maybe_refine_async hard timeout + status; REST polish parity + refined_text
polish; warmup builds the recognizer and a second session reuses it; the panel
honesty note. Backend refinement/capture_ws/capture/sherpa suites, CJK + route
inventory gates, full vitest (733), lint (0 errors) and format all green.
Co-authored-by: mergetest <test@local>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* feat(dictation): opt-in NLMS AEC for dictate-over-playback (Wave 8b)
Dictating while OmniVoice plays audio (TTS preview, dub, video) leaks the
loudspeaker signal into the mic, and the streaming ASR transcribes that
bleed. Browser echoCancellation varies per platform/webview — it can't be a
cross-platform default — so this adds a server-side canceller that behaves
identically everywhere.
services/aec.py ports Patter's NlmsEchoCanceller (MIT): a time-domain NLMS
adaptive filter with a Geigel double-talk detector, warm-up step ramp, and
far-end staleness pass-through. /ws/transcribe gains an opt-in '?aec=1[&sr=]'
mode: frames are raw int16 mono PCM tagged with a 1-byte prefix (0x00 mic,
0x01 playback reference); the mic is cleaned against the reference before
buffering, and the cleaned PCM is muxed via stdlib wave (not ffmpeg). Without
the param the protocol and behaviour are byte-for-byte unchanged.
Backend ships dark (no new deps — numpy already pinned); frontend far-end
streaming is a follow-up. Tests cover echo attenuation, double-talk
preservation, cold/stale pass-through, param validation, and the framing
helpers — all pure-numpy/stdlib so they skip the torch ASR stack.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(capture_ws): stubs accept the new pcm_sr kwarg
_transcribe_buffer/_transcribe_buffer_full gained an optional pcm_sr kwarg
for the AEC PCM path; the protocol-test stubs had fixed signatures and
raised TypeError on it, so the handler sent 'error' instead of 'final'.
Accept **kw in the stubs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Tray + lifecycle:
- tauri-plugin-single-instance — second launch focuses existing window
instead of racing for port 3900.
- Window close hides instead of destroying; backend shutdown moved to
RunEvent::ExitRequested so only the tray "Quit" item (or Cmd+Q on macOS)
actually exits.
- Tray icon flips to red-dot variant during dictation recording.
Hotkey customization:
- Settings → Capture tab. Records any modifier+key combo, persists to
app config, re-registers on launch.
- set_dictation_shortcut rolls back to the previous binding on register
failure so a bad combo never leaves the user with no shortcut.
Dictation latency / correctness:
- WS-final treated as source of truth; HTTP POST /transcribe runs only as
fallback (WS error / timeout / no-WS path). Audio transcribed once
instead of twice. Server accepts an "EOF" text frame (or empty binary
frame) so the socket stays open for `final` to be delivered before the
client closes.
- MediaRecorder chunks queued during the WS handshake are drained in
ws.onopen — the server's final transcript no longer drops the first
~250 ms of audio.
- Fallback timeout scales with recording length (max(15s, recordedMs+10s))
so long-form dictations don't trip duplicate transcription.
Donate page:
- Drop Patreon, Bitcoin / Ethereum / Solana cards. Drop qrcode.react.
- Move "Commercial License" CTA from page bottom to top-right header bar.
Docker hygiene:
- docker-compose binds 127.0.0.1 by default. README documents the LAN
exposure trade-off + recommends a reverse proxy with auth.
CI:
- New cross-platform `tauri-cross-platform` job runs `cargo check` against
the Tauri shell on macOS / Windows / Linux per PR. Catches platform
cfg-gate regressions without paying the full ~15min/platform bundle
cost (full bundling stays in release.yml on tag push).
Tests:
- tests/test_capture_ws.py (3 cases) covers EOF text-frame, empty-binary
EOF, and legacy disconnect-finalize paths.
Includes the user's previously-staged 0.2.5 polish: cross-platform
desktop-prod.sh, Dockerfile base-image fix, bun.lock churn.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Palash Debnath