talktype

Push-to-talk speech-to-text for Linux. Press a hotkey to start recording, press it again to transcribe and type the text wherever your cursor is. No GUI, no app to keep running — just a keyboard shortcut.

• Pluggable backends — swap transcription models without changing anything else
• Works everywhere — GNOME, Sway, Hyprland, i3, X11
• ~160 lines of bash — easy to read, easy to hack on

Ships with faster-whisper by default, plus an optional Moonshine backend for CPU. Or bring your own — anything that reads a WAV and prints text works.

Note: This project is in early development — expect rough edges. If you run into issues, please open a bug.

Requirements

• Linux (Wayland or X11)
• Audio recorder: ffmpeg (preferred) or PipeWire (pw-record)
• Typing tool (auto-detected, best available is used):
  • wtype — Wayland (Sway, Hyprland; not GNOME)
  • ydotool + ydotoold — Wayland & X11 (preferred with daemon)
  • xdotool — X11 only (not Wayland)
  • ydotool without daemon — last resort, with warning
• socat (for server-backed transcription)

For the default backend (faster-whisper):

• NVIDIA GPU with CUDA (or use CPU mode — see Whisper backend options)

Install

git clone https://github.com/csheaff/talktype.git
cd talktype
make install

This will:

1. Install system packages (wtype, ydotool, etc.)
2. Create a Python venv with faster-whisper
3. Symlink talktype into ~/.local/bin/

ydotool permissions

Note: Only needed if you use ydotool. If you use wtype (Wayland) or xdotool (X11), skip this.

ydotool needs access to /dev/uinput. Add yourself to the input group:

sudo usermod -aG input $USER
echo 'KERNEL=="uinput", GROUP="input", MODE="0660"' | sudo tee /etc/udev/rules.d/80-uinput.rules
sudo udevadm control --reload-rules && sudo udevadm trigger

Then reboot for the group change to take effect.

Pre-download model (optional)

make model

Configuration

talktype reads ~/.config/talktype/config on startup (follows $XDG_CONFIG_HOME). This works everywhere — GNOME shortcuts, terminals, Sway, cron — no need to set environment variables in each context.

mkdir -p ~/.config/talktype
cat > ~/.config/talktype/config << 'EOF'
TALKTYPE_CMD="/path/to/talktype/transcribe-server transcribe"
EOF

Any TALKTYPE_* variable can go in this file. Environment variables still work and are applied after the config file, so they override it.

Set TALKTYPE_TYPE_CMD to control which typing tool is used (auto, wtype, ydotool, xdotool, or any custom command). Default is auto, which picks the best available tool: wtype (Wayland) → ydotool+daemon → xdotool (X11).

Setup

Bind talktype to a keyboard shortcut:

GNOME: Settings → Keyboard → Keyboard Shortcuts → Custom Shortcuts

• Name: TalkType
• Command: talktype (or full path ~/.local/bin/talktype)
• Shortcut: your choice (e.g. Super+D, F11, etc.)

Sway / Hyprland: Add to your config:

bindsym $mod+d exec talktype

Usage

1. Press your shortcut → notification says "Listening..."
2. Speak
3. Press the shortcut again → transcribes and types the text at your cursor

Backends

Server backends auto-start on first use — the model loads once and stays in memory for fast subsequent transcriptions.

Whisper (default)

faster-whisper. Best with a GPU. Works out of the box after make install with no config needed.

For faster repeated use, switch to server mode in your config:

# ~/.config/talktype/config
TALKTYPE_CMD="/path/to/talktype/transcribe-server transcribe"

Variable	Default	Description
WHISPER_MODEL	base	tiny, base, small, medium, large-v3-turbo
WHISPER_LANG	en	Language code
WHISPER_DEVICE	cuda	cuda or cpu
WHISPER_COMPUTE	float16	float16 (GPU), int8 or float32 (CPU)

Moonshine (CPU, lightweight)

Moonshine by Useful Sensors. 61.5M params, purpose-built for CPU/edge inference.

make moonshine

# ~/.config/talktype/config
TALKTYPE_CMD="/path/to/talktype/backends/moonshine-server transcribe"

Set MOONSHINE_MODEL=UsefulSensors/moonshine-tiny for an even smaller 27M param model.

Server management

The server starts automatically on first transcription. You can also manage it directly:

./transcribe-server start   # start manually
./transcribe-server stop    # stop the server

Custom backends

Set TALKTYPE_CMD to any command that takes a WAV file path as its last argument and prints text to stdout:

# ~/.config/talktype/config
TALKTYPE_CMD="/path/to/my-transcriber"

Your command will be called as: $TALKTYPE_CMD /path/to/recording.wav

It should print the transcribed text to stdout and exit. That's the only contract — use whatever model, language, or runtime you want.

How it works

[hotkey] → recording starts → [hotkey] → recording stops
                                            ↓
                                     $TALKTYPE_CMD audio.wav
                                            ↓
                                     type_text → text appears at cursor

The talktype script is ~160 lines of bash. Transcription backends are swappable. Server mode uses Unix sockets to keep models in memory.

License

MIT