VoiceInput

A fast, lightweight voice dictation tool for Linux (Wayland) built in C. Supports both Google Gemini and self-hosted Ollama models.

Features

• Gemini 2.5 Flash Lite - Native audio transcription via Google Cloud API
• Ollama - Self-hosted whisper-tiny + gpt-oss:cloud
• Global hotkey trigger - Works with Niri compositor keybindings
• FIFO-based triggering - External scripts can trigger dictation
• Text injection - Uses wtype for Wayland-compatible text input
• Transcription history - SQLite-backed local storage

Requirements

Arch Linux Packages

sudo pacman -S portaudio curl cjson libinput libei sqlite cmake gcc make pkgconf wtype

Building

cd /home/antony/coding/voiceinput
mkdir build && cd build
cmake ..
make

Configuration

1. Set your Gemini API key

Edit ~/.config/voiceinput/config.json:

{
  "provider": "gemini",
  "gemini": {
    "api_key": "YOUR_GEMINI_API_KEY_HERE"
  }
}

Get an API key from: https://aistudio.google.com/apikey

2. Configure Niri keybinding

Add to your Niri config (~/.config/niri/config.kdl):

keybinds = [
  {
    keys = ["Space"]
    action = { spawn = ["/home/antony/coding/voiceinput/trigger.sh"] }
  }
]

Then reload Niri config (usually Super+Shift+R or restart).

3. Full config options

{
  "provider": "gemini",
  "ollama": {
    "host": "http://localhost:11434",
    "transcription_model": "whisper-tiny",
    "processing_model": "gpt-oss:cloud"
  },
  "gemini": {
    "api_key": "YOUR_API_KEY"
  },
  "hotkey": {
    "modifier": "Ctrl",
    "key": "Space"
  },
  "audio": {
    "sample_rate": 16000,
    "channels": 1,
    "chunk_duration_ms": 5000
  },
  "text_injection": {
    "method": "wtype",
    "fallback": "wtype"
  },
  "refinement": {
    "enabled": true,
    "remove_fillers": true,
    "auto_punctuate": true
  },
  "history": {
    "enabled": true,
    "max_entries": 1000
  }
}

Usage

Method 1: Niri Keybinding (Recommended)

1. Add the keybinding to Niri config (see above)
2. Run voiceinput in background:

   ./build/voiceinput &

3. Press Space to dictate - text appears at cursor!

Method 2: Manual Trigger

# Start voiceinput
./build/voiceinput &

# Trigger from anywhere
echo "trigger" > /tmp/voiceinput-trigger

Method 3: Hotkey (Fallback)

If libinput works in your setup:

• Press Ctrl+Space to start dictation
• Release to stop and transcribe

Architecture

┌─────────────────────────────────────────────────────────┐
│  Trigger (Niri keybind / FIFO / libinput hotkey)        │
├─────────────────────────────────────────────────────────┤
│  Audio Capture (PortAudio, 16kHz mono)                  │
├─────────────────────────────────────────────────────────┤
│  Transcription Provider                                 │
│    - Gemini 2.5 Flash Lite (native audio)               │
│    - Ollama (whisper-tiny)                              │
├─────────────────────────────────────────────────────────┤
│  AI Refinement (Gemini or gpt-oss:cloud)                │
├─────────────────────────────────────────────────────────┤
│  Text Processor (filler removal, punctuation)           │
├─────────────────────────────────────────────────────────┤
│  Text Injector (wtype for Wayland)                      │
├─────────────────────────────────────────────────────────┤
│  History Storage (SQLite)                               │
└─────────────────────────────────────────────────────────┘

Troubleshooting

Gemini API errors

• Check your API key is set in config
• Verify network connectivity: curl https://generativelanguage.googleapis.com

Text not appearing

• Ensure wtype is installed: which wtype
• Test wtype manually: echo "test" | wtype -

Trigger not working

• Check FIFO exists: ls -la /tmp/voiceinput-trigger
• Ensure voiceinput daemon is running

Audio not captured

• Check microphone permissions
• Test with: arecord -d 2 | aplay

Project Structure

voiceinput/
├── src/
│   ├── main.c           # Entry point, event loop, FIFO trigger
│   ├── audio.c          # PortAudio audio capture
│   ├── ollama.c         # Ollama API client
│   ├── gemini.c         # Gemini API client (native audio)
│   ├── hotkey.c         # libinput hotkey listener
│   ├── inject.c         # Text injection (wtype)
│   ├── textproc.c       # Text processing/cleanup
│   ├── config.c         # JSON configuration parser
│   └── history.c        # SQLite history storage
├── include/
│   └── voiceinput.h     # Common headers/types
├── trigger.sh           # Niri trigger script
├── niri-config.kdl      # Example Niri config snippet
├── CMakeLists.txt       # Build configuration
├── config.example.json  # Example configuration
└── README.md            # This file

License

MIT License

Acknowledgments

Inspired by:

• Wispr Flow - Voice dictation with AI editing
• Voquill - Open-source voice typing