voculinux/docs/USER_GUIDE.md

User Guide

This guide explains how to use Vocalinux effectively.

Getting Started

After installing Vocalinux (see the Installation Guide), you can start the application from the terminal and optionally enable start-on-login.

Start on Login (Autostart)

Vocalinux supports login autostart using the standard Linux desktop-session mechanism.

• Where to enable it:
  • First-run welcome dialog
  • Tray menu: Start on Login
  • Settings dialog: Start on Login
• What Vocalinux creates:
  • vocalinux.desktop in $XDG_CONFIG_HOME/autostart/ or ~/.config/autostart/
• How it starts:
  • As a regular user GUI app in your desktop session (--start-minimized)
• What it does not do:
  • It does not create a systemd service/unit for autostart

Desktop Compatibility Notes

• Works on most mainstream desktop environments (GNOME, KDE, Xfce, Cinnamon, MATE, LXQt)
• On minimal/custom window managers, autostart may require an autostart manager or desktop-specific startup hook

Basic Usage

Starting and Stopping Voice Typing

1. Launch the application: Run vocalinux in a terminal or launch it from your application menu
2. Find the tray icon: Look for the microphone icon in your system tray
3. Start voice typing: Hold Right Alt (Option) by default, or use the tray menu / your configured shortcut
4. Speak clearly: As you speak, your words will be transcribed into the currently focused application
5. Stop voice typing: Release the key in push-to-talk (default), double-tap again in toggle mode, or use the tray menu

Dictation formatting

Vocalinux capitalizes the start of dictation and letters after ., !, or ?. Each completed utterance also leaves a trailing space so the next push-to-talk or toggle session does not glue onto the previous sentence (Hello.This → Hello. This).

Dictating into terminals

When Vocalinux injects through the clipboard (the usual Wayland / ydotool path), it sends Ctrl+V in ordinary text fields and Ctrl+Shift+V in terminal emulator windows. Auto-detect works on X11 and on Hyprland, Sway, and niri. On GNOME or KDE Wayland, set Settings → Dictation → Clipboard Paste Shortcut to Ctrl+Shift+V.

On non-US layouts such as German Neo, that chord uses the key that types v on the active layout (not physical KEY_V). Plasma's active layout comes from layout memory or D-Bus, not from kxkbrc list order. The map is cached for the process lifetime — restart Vocalinux after switching layouts. Per-window Plasma layouts may be stale if D-Bus is unavailable.

Nested terminal panels inside an IDE are often invisible to window-class detection. If paste lands as a literal ^V or does nothing, open Settings → Dictation → Clipboard Paste Shortcut and choose Ctrl+Shift+V. Choose Ctrl+V if a window was mis-detected as a terminal.

Understanding the Status Icons

• Microphone off (gray): Voice typing is inactive
• Microphone on (blue): Voice typing is active and listening
• Microphone processing (orange): Voice typing is processing your speech

Voice Commands

Vocalinux supports several commands that you can speak to control formatting. English phrases always work. When recognition language is set (e.g. Italian, French, German, Spanish), matching punctuation and line-break phrases in that language are also recognized (for example Italian virgola / punto, French virgule / point).

Command	Action
"new line" or "new paragraph"	Inserts a line break
"period", "full stop", or "dot"	Types a period (.)
"comma"	Types a comma (,)
"question mark"	Types a question mark (?)
"exclamation point" or "exclamation mark"	Types an exclamation point (!)
"semicolon"	Types a semicolon (;)
"colon"	Types a colon (:)
"delete that" or "scratch that"	Deletes the last sentence
"capitalize" or "uppercase"	Capitalizes the next word
"all caps"	Makes the next word ALL CAPS

Editing and formatting action phrases (delete that, undo, capitalize, …) are currently English-only.

Tips for Better Recognition

1. Use a good microphone: A quality microphone significantly improves recognition accuracy
2. Speak clearly: Enunciate your words clearly but naturally
3. Moderate pace: Don't speak too quickly or too slowly
4. Quiet environment: Minimize background noise when possible
5. Learn commands: Familiarize yourself with voice commands for punctuation and formatting
6. Use GPU acceleration: If you have a GPU (AMD, Intel, or NVIDIA), whisper.cpp will automatically use it for faster transcription
7. Choose the right model:
   • For real-time dictation: Use tiny or base (fastest)
   • For better accuracy: Use small, medium, or large
   • For English-only dictation: Choose an .en specialization
   • For lower-memory systems: Choose a quantized specialization such as q5_0 or q5_1
8. Check debug logs: Run vocalinux --debug to see which backend is being used (Vulkan, CUDA, or CPU)

Customization

Keyboard Shortcut

Vocalinux supports two shortcut modes for controlling voice typing:

• Push-to-talk mode (default): Hold Right Alt (Option on Mac-layout keyboards) to speak, then release to stop
• Toggle mode: Double-tap the configured shortcut key to start/stop voice typing
• Configure mode and key in Settings -> Shortcuts

Model Settings

You can change the speech recognition engine and model for better accuracy or faster performance:

Choosing Your Engine

Vocalinux now offers three speech recognition engines:

1. whisper.cpp ⭐ (Default) - High-performance C++ engine

   • Fastest installation (~1-2 min)
   • Works with AMD, Intel, NVIDIA GPUs via Vulkan
   • True multi-threading (no Python GIL)
   • Best for most users

2. Whisper (OpenAI) - PyTorch-based engine

   • NVIDIA GPU only (requires CUDA)
   • Larger download (~2.3GB with PyTorch)
   • Installation takes ~5-10 min
   • Use if you specifically need PyTorch features

3. VOSK - Lightweight engine

   • Smallest footprint (~40MB)
   • CPU only
   • Great for older systems or minimal resource usage

Changing Engine and Model

1. Open settings from the tray icon menu (right-click)
2. Open the Speech Engine page in the settings sidebar (search works if you prefer)
3. Select your Speech Engine:
   • whisper_cpp (recommended)
   • whisper
   • vosk
4. Select your Model Size:
   • tiny (~74MB) - Fastest, good for real-time dictation
   • base (~141MB) - Good balance
   • small (~465MB) - Better accuracy
   • medium (~1.5GB) - High accuracy
   • large (~3.0GB) - Best accuracy, slower
5. For whisper.cpp, select a Specialization:
   • Standard multilingual - Best default for auto-detect or non-English dictation
   • English-only - Choose when you dictate only in English
   • Quantized - Lower memory and smaller downloads with a possible accuracy tradeoff
   • Turbo - Faster large-v3 option with strong accuracy
   • Legacy large - Use only if you specifically need an older large model version

English-only whisper.cpp specializations limit the language selector to English.

Removing unused models

Open Settings and go to Speech Model. If leftover files are on disk that are not the model currently selected, Unused downloads appears under the model info card. Expand it to delete leftovers one at a time. Confirming removes those files from ~/.local/share/vocalinux. Vocalinux will download a model again if you select it later. Packaged system-wide VOSK models are left alone. The section is hidden when there is nothing unused to delete.

When to Use Each Model

For real-time dictation: Use tiny or base - they're fast enough to keep up with your speech.

For transcription: Use small or medium - better accuracy for recorded audio.

For maximum accuracy: Use large - best results but requires more RAM and GPU power.

For lower-memory systems: Use a quantized whisper.cpp specialization such as Q5 or Q8.

For English-only dictation: Use an English-only specialization and keep the language set to English.

GPU Acceleration

whisper.cpp uses GPU acceleration when the bundled pywhispercpp libraries include Vulkan or CUDA:

• Vulkan (AMD, Intel, NVIDIA) when libggml-vulkan is present
• CUDA (NVIDIA) when libggml-cuda is present
• CPU when those libraries are missing, even if Settings lists a GPU

Host tools such as vulkaninfo only describe the machine. The engine follows the libraries actually loaded. Look for GPU backend: vulkan or GPU backend: cuda after the model loads. CPU-only; pywhispercpp lacks GPU libraries means inference is on the CPU.

On multi-GPU machines, Vocalinux prefers a discrete Vulkan device when one is present and skips software renderers such as llvmpipe. Override the device under Settings → Performance → Vulkan GPU.

Pip wheels of pywhispercpp are often CUDA builds. In that case Vocalinux always uses CUDA device 0 (the first NVIDIA GPU), because Vulkan GPU indices do not match CUDA ordinals. On a hybrid laptop the NVIDIA GPU is usually Vulkan GPU 1; feeding that index into CUDA would land on the CPU fallback instead.

If you need a second NVIDIA GPU, either set CUDA_VISIBLE_DEVICES before starting Vocalinux or install a Vulkan-built pywhispercpp so the Performance GPU picker applies.

install.sh rebuilds pywhispercpp with Vulkan or CUDA when it can. Prefer that launcher (or the wrapper it installs) so LD_LIBRARY_PATH includes the GPU libs. Raw venv/bin binaries can come up CPU-only.

AppImages built from this tree compile pywhispercpp with Vulkan and use the host Vulkan driver at runtime. Older AppImages shipped the CPU-only wheel; if you see the lacks GPU libraries line, download a newer AppImage or use install.sh.

Hybrid / PRIME / Optimus laptops: picking NVIDIA in Settings is enough when pywhispercpp is Vulkan-built. prime-run is not required for Vulkan, and it cannot fix a CPU-only wheel. Install vulkan-tools so vulkaninfo can enumerate devices; without it, whisper.cpp may default to GPU 0 (often the iGPU).

To check which backend is being used, look for these log messages when starting Vocalinux (vocalinux --debug):

[INFO] whisper.cpp using Vulkan GPU backend: AMD Radeon RX 6800
[INFO] Using Vulkan GPU [0]: AMD Radeon RX 6800
[INFO] whisper.cpp configured with n_threads=4 (GPU backend: vulkan)

Auto-pause and model keep-alive

Optional power-saving controls live under settings:

• Auto-pause apps — unload the speech model while configured apps/games are running, then reload when they exit
• Model keep-alive — unload the model after a configurable idle timeout so idle dictation does not keep GPU/CPU resources warm

Troubleshooting

If you encounter issues, check the Installation Guide troubleshooting section or run the application with debug logging:

vocalinux --debug

Check the logs for error messages and possible solutions.