Skip to content

Latest commit

 

History

History
101 lines (66 loc) · 5.08 KB

File metadata and controls

101 lines (66 loc) · 5.08 KB

Troubleshooting

Docs: Configuration · Troubleshooting


Hotkey Not Working

  • Ensure no other app is using the same hotkey combination.
  • Some apps (Discord overlays, GeForce Experience, antivirus software) intercept hotkeys at a lower level and will silently win. Try a less common combination like Ctrl+Shift+Space or Alt+F9.
  • The Logs tab will show a "Hotkey registered" message on startup — if it's missing, the listener failed to start.
  • On Windows, hotkeys cannot be captured from windows running at a higher privilege level (e.g. Task Manager). If your target window is elevated, try running LogNotes as administrator.

Text Not Pasting

  • Make sure a text input field is focused before releasing the hotkey — the app waits 0.5s for focus to return, but some apps take longer.
  • Some elevated windows (Task Manager, UAC prompts, certain terminals) block simulated keystrokes from non-elevated processes. Run LogNotes as administrator to paste into them.
  • The clipboard is automatically cleared 5 seconds after pasting for security — this is intentional. If you need to paste the same text again, use the Copy button in the Activity tab.
  • Check the Logs tab for paste_text success or error messages to confirm whether the paste was attempted.

No Audio Recorded

  • Check Windows Settings → System → Sound → Input to confirm the correct microphone is selected as the default input device.
  • Check Windows Settings → Privacy & Security → Microphone — LogNotes must have microphone access enabled.
  • Speak at a normal volume; very quiet input may not register.
  • The Logs tab will show "No audio recorded" if the recording captured complete silence, or an audio backend error if the microphone could not be opened.

First Transcription Is Slow

  • The Whisper model loads on first use (~2-5 seconds). Status will show "Loading models…" during this time.
  • After the first load, all three Whisper sizes are warmed in the background so subsequent model switches are instant.
  • This is a one-time cost per session — subsequent transcriptions are fast.

Transcription Quality Is Poor

  • Switch to a larger model: base → small. small offers the best quality/speed tradeoff on CPU.
  • Speak closer to the microphone and reduce background noise.
  • Speak at a natural pace — very fast or very slow speech can confuse the model.

Overlay in Wrong Position / Hidden Behind Taskbar

  • Right-click the overlay to cycle through the four corners.
  • Left-click and drag to reposition manually.
  • The corner is persisted in config; the exact drag position is not — it resets to the configured corner on restart.
  • The overlay is positioned within the primary display's work area to avoid the taskbar, but secondary monitors with non-standard taskbar positions may still cause overlap. Switch corners to work around it.

Retry Button in Activity Tab Doesn't Respond

  • The app is still processing — clicks are dropped while busy. The status bar will show "Busy — retry again when ready".
  • Selecting a model you have never used before triggers a one-time download from HuggingFace (up to several minutes on slow connections). Check the Logs tab for a "Retry starting…" entry to confirm it is running.
  • Subsequent retries with the same model load from the local cache in a few seconds.

High memory usage

  • Torch and the Whisper models are large. With all three sizes warmed, expect ~1–2 GB RAM usage — this is normal.
  • If memory is constrained, restart the app to clear warmed models, and only use one Whisper size.

Config File Corrupted or Settings Behaving Unexpectedly

  • Delete %APPDATA%\LogNotes\config.json and restart the app — it will be recreated with defaults.
  • If you edit the file manually, ensure it is valid JSON. Invalid JSON causes the app to fall back to all defaults on next launch (an error is logged).
  • See configuration.md for valid values per key.

SmartScreen Warning on First Launch (Packaged Build)

  • The unsigned build triggers "Windows protected your PC" — this is expected.
  • Click More info → Run anyway to proceed.
  • This warning only appears once per build. Code signing (~$100/yr OV cert) would eliminate it if distributing publicly.

Packaged App Opens but Stays "Connecting to Sidecar…"

The window appears but never reaches "Ready" — the Electron front end can't reach the Python back end.

  • Run the bundled back end directly to see its error. It is built with a console, so it prints what fails:
    "%LOCALAPPDATA%\Programs\LogNotes\resources\sidecar\LogNotes.exe"
    (Path varies with the install location; look under the install dir's resources\sidecar\.) A healthy run prints a PORT <n> line.
  • Common causes: a missing DLL / hidden import, or a sounddevice / PortAudio payload issue. The console output identifies the failing module.
  • First launch is slow (models load); give it time before assuming it is stuck.
  • Make sure no orphaned LogNotes.exe from a previous run is interfering (Task Manager).