Smart Voice Dictation on Linux: Automatically Detect Keyboard Layout

Imagine you’re writing a document in both English and Arabic. You need to switch between dictation modes every time you change languages. It’s tedious. On Windows, you just press Win+H and start talking. But on Linux, you’re stuck manually selecting the right speech model.

That’s the problem I set out to solve.

The Solution

I created nerd-dictation-auto-switch-languages, a wrapper script that automatically detects your current keyboard layout and uses the appropriate speech-to-text model. No more manual switching. Just press one shortcut and start talking.

How It Works

1. Press Super+H to start dictation
2. The script detects your keyboard layout (English, Arabic, German, etc.)
3. Launches the correct VOSK model automatically
4. Speak — your words appear on screen
5. Press Super+Shift+H to stop

The Technical Details

Key Components

The Magic: Layout Detection

The key innovation is detecting the keyboard layout in real-time:

# Get current keyboard layout
CURRENT_LAYOUT=$(xkblayout-state print "%s")

# Result: "us", "ara", "de", etc.

Then map it to the appropriate model:

case "$CURRENT_LAYOUT" in
    us)
        MODEL_DIR="$HOME/.config/nerd-dictation/model"
        LANG_NAME="English"
        ;;
    ara)
        MODEL_DIR="$HOME/.config/nerd-dictation/model-ar"
        LANG_NAME="Arabic"
        ;;
esac

The Scripts

dictate-start — Main script that:

• Detects keyboard layout
• Selects appropriate model
• Starts dictation with 30-second timeout
• Shows desktop notification

dictate-stop — Cleanup script that:

• Reads which model was used
• Stops the dictation process
• Types the transcribed text
• Shows completion notification

Installation

Prerequisites

# Install system dependencies
sudo apt install python3-pip xdotool zenity libnotify-bin wget unzip git

# Install xkblayout-state
git clone https://github.com/nonpop/xkblayout-state.git
cd xkblayout-state && make && sudo make install

Download Models

# English (required)
mkdir -p ~/.config/nerd-dictation
wget https://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip
unzip vosk-model-small-en-us-0.15.zip
mv vosk-model-small-en-us-0.15 ~/.config/nerd-dictation/model

# Arabic (optional)
wget https://alphacephei.com/vosk/models/vosk-model-ar-mgb2-0.4.zip
unzip vosk-model-ar-mgb2-0.4.zip
mv vosk-model-ar-mgb2-0.4 ~/.config/nerd-dictation/model-ar

Set Up Scripts

# Clone scripts
cp dictate-start ~/.nerd-dictation/
cp dictate-stop ~/.nerd-dictation/
chmod +x ~/.nerd-dictation/dictate-*

Configure Shortcuts

In your desktop settings:

Results

Before

1. Switch keyboard to Arabic
2. Find and open Arabic dictation app
3. Start dictation
4. Speak
5. Switch back to English
6. Find and open English dictation app
7. Repeat...

After

1. Press Super+H
2. Speak
3. Press Super+Shift+H
4. Done!

Supported Languages

Adding new languages is straightforward — just download the VOSK model and add a case to the script.

Challenges & Solutions

Challenge 1: Reliable Layout Detection

Problem: gsettings didn’t update in real-time when switching layouts.

Solution: Used xkblayout-state which reads directly from X11.

Challenge 2: Model Organization

Problem: Multiple language models needed management.

Solution: Standardized directory naming (model, model-ar, model-de).

Challenge 3: Error Handling

Problem: Users might not understand why dictation fails.

Solution: Desktop notifications + zenity popups with clear messages.

Future Improvements

• Add support for more languages
• Implement auto-language switching mid-dictation
• Create GUI for configuration
• Add Whisper as alternative backend

Conclusion

Linux speech recognition has come a long way. With nerd-dictation-auto-switch-languages and VOSK, you can have seamless, offline, private voice typing that automatically adapts to your language. No cloud services, no subscriptions, just open-source speech-to-text that just works.

The project is fully open-source and available on GitHub. Contributions welcome!

Resources

• nerd-dictation-auto-switch-languages Repository
• nerd-dictation
• VOSK Models
• xkblayout-state

Have questions or suggestions? Open an issue on GitHub!