# VoZii Troubleshooting Where to look first: the **log file**. Its location depends on your install: | Install type | Log location | | --- | --- | | Normal | `%LOCALAPPDATA%\VoZii\vozii.log` | | Portable (a `config.yaml` / `whisper-cpp` folder sits next to `VoZii.exe`) | `vozii.log` next to the `.exe` | | App crashed before any window appeared | `%TEMP%\vozii_boot_error.log` | Open it quickly from the tray menu: **right-click the VoZii tray icon → Open log**. --- ## Recording works, but no text appears The most common causes are Windows blocking the (unsigned) speech engine, or a missing system runtime. Since v1.8 VoZii detects both and shows a dialog with the exact reason — but here is the background: ### Smart App Control blocks the speech engine **Symptom:** the settings window opens, recording starts, but every dictation comes back empty. The log shows: ``` whisper-server sofort beendet (Code 3236495362) ``` `3236495362` = `0xC0E90002` — Windows **Smart App Control** (SAC) refuses to load the unsigned `ggml-*.dll`. You can confirm it in Event Viewer under *Applications and Services → Microsoft → Windows → CodeIntegrity → Operational* (event **3077**, policy `VerifiedAndReputableDesktop`). **Fix:** Windows Security → **App & browser control** → **Smart App Control** → **Off**. > ⚠️ SAC is a one-way switch: once turned off, it can only be re-enabled by > reinstalling Windows. SAC also has **no per-app exceptions**, and Defender > exclusions do not help — turning it off (or a signed VoZii build in the > future) are the only options. ### Missing VC++ Runtime (fresh Windows) **Symptom:** on a freshly installed Windows, dictation stays empty; the log shows exit code `3221225781` (= `0xC0000135`, *DLL not found*). The whisper.cpp binaries need the **Microsoft Visual C++ 2015–2022 Redistributable (x64)**. VoZii checks for it at startup and shows a download dialog; you can also install it directly: --- ## The app does not start at all - **SmartScreen** ("Windows protected your PC"): expected for an unsigned `.exe` — click *More info → Run anyway*, or right-click the file → *Properties → Unblock → Apply*. Verify your download against the `SHA256SUMS.txt` attached to each release. - **Nothing happens, no window:** VoZii is a onefile `.exe` that unpacks itself into `%TEMP%\_MEI...` at launch. Corporate policies (AppLocker, WDAC) or aggressive antivirus can block execution from `%TEMP%`. Check `%TEMP%\vozii_boot_error.log` and your AV quarantine, and allow the app. - **"VoZii is already running":** check the tray (bottom-right, behind the ^ arrow). If no icon is there, end a stale `VoZii.exe` in Task Manager. --- ## It worked for hours, then stopped (pre-1.9 issue) Four separate causes were found and fixed in **v1.9.0**. If you are on an older build, update — the workaround for all of them was restarting VoZii. | Symptom in the log | Cause | | --- | --- | | `PyAutoGUI fail-safe triggered` | The mouse pointer sat in a screen corner; the old input library aborted every keystroke there. Text stayed on the clipboard. | | Nothing at all after `Aufnahme gestartet` | The audio stream had gone silent (USB mic re-enumerated, Windows audio engine restarted) without PortAudio reporting it. | | The *previous* clipboard content was pasted | The clipboard was restored before the target app had read it. | | `Zwischenablage ... belegt` (v1.9+) | Another app (Office, Teams, RDP, a clipboard manager) held the clipboard. VoZii now retries for up to a second and refuses to paste rather than paste the wrong text. | Since v1.9.0 a watchdog checks the hotkey hooks and the audio stream every 15 seconds and repairs both on its own, and a dead microphone shows up as `ERR:MIC` in the overlay instead of doing nothing. ## Microphone issues - Use the built-in **Test** button (settings → Microphone) — a live level meter confirms the signal. - If your configured microphone was unplugged, VoZii automatically falls back to the system default and logs a warning. - USB microphones sometimes re-enumerate after standby; VoZii reopens the stream on the next recording automatically — and since v1.9.0 also while idle, so the next hotkey press records again without a restart. ## Model / download issues - Downloads resume automatically after a connection loss — just press the button again. - Every download is verified against a pinned SHA-256 checksum; a corrupted file is discarded automatically. Retry, and if it persists, open an issue. - To force a fresh model download: delete the model file under `whisper-cpp\models\` and press **Download** again. - After a GPU/backend change (e.g. new graphics card), the settings window offers an **Update** button that reinstalls the matching whisper.cpp build. ## Slow or delayed transcription - VoZii keeps a `whisper-server` in RAM (the model stays loaded). If that fails, it silently falls back to a slower per-call CLI mode — the log tells you which engine is active (`Transcriber-Engine: whisper (server)`, `whisper (cli)` or `parakeet (cpu)`). - **No graphics card?** Switch the engine to **Parakeet (CPU)** in the model section. It runs ~20x faster than real time on ordinary laptop CPUs (a 10-second dictation takes about half a second) and covers 25 European languages. Measure it on your own machine with `python scripts/bench_stt.py your.wav --whisper large-v3-turbo-q5_0`. - With Whisper on weak CPUs prefer the **Recommended** (turbo q5) model over *Best*, and the **Fast** decoding mode. - Overlay shows **`GPU ERROR → CPU`**: the CUDA/Vulkan backend crashed (driver, VRAM). VoZii switches to the CPU for the rest of the session — Parakeet if installed, otherwise whisper.cpp with `--no-gpu`. Updating the graphics driver usually fixes the root cause.