stttype 3.0.5

Cross-platform voice-to-text typing assistant with GPU acceleration and automatic CPU fallback

STT Type v3.0.0

Cross-platform voice-to-text typing assistant with real-time streaming transcription.

Hold F2 (configurable) to stream your voice in real-time, release to type the final transcribed text at your cursor position.

What's New in v3.0.0

• Real-time streaming mode — Uses WhisperLive for continuous transcription while you speak. Interim results appear in the console in real-time.
• Hold mode (legacy) — Still available! Record on hold, transcribe on release using faster-whisper directly.
• Mode selector — Choose between Streaming and Hold in stttype --config
• Fully offline — No internet required after initial model download

Features

• Real-time streaming — See transcription progress as you speak (Streaming mode)
• Cross-platform — Works on Windows, Linux, and macOS
• Hold hotkey to record/stream — Audio captures while key is held (default: F2)
• Visual indicator — Transparent red dot with hotkey label appears in top-right corner while recording
• Bell sounds — Audio feedback when recording starts/stops
• GPU-accelerated STT — Uses faster-whisper on your NVIDIA GPU
• Auto CPU fallback — Automatically falls back to CPU if GPU is not available
• Clipboard typing — Types text via Ctrl+V paste for reliability (no key interference)
• System tray mode — Runs silently in background
• Auto-startup — Starts automatically on login (Windows)
• Settings GUI — stttype --config to change model, language, device, compute type, hotkey, and mode

Requirements

• Python 3.9+
• Microphone
• NVIDIA GPU with CUDA support (optional, for GPU mode)
• Linux: paplay, aplay, or pw-play for sound feedback (auto-detected)

Installation

From PyPI (Recommended)

pip install stttype

Prerequisites (Optional GPU Support)

Install PyTorch with CUDA support for GPU acceleration:

# Windows/Linux with CUDA 11.8
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118

# macOS (CPU only, no CUDA)
pip install torch torchaudio

Without PyTorch, STT Type will still work — it automatically falls back to CPU.

Windows (Local Install)

cd "E:\Lucas\STT Type"
.\install.ps1

Then restart PowerShell.

Linux

cd /path/to/stttype
chmod +x install.sh
./install.sh

If sounddevice fails, install PortAudio:

# Debian/Ubuntu
sudo apt-get install portaudio19-dev

# Fedora
sudo dnf install portaudio-devel

# Arch
sudo pacman -S portaudio

For sound feedback, ensure one of these is installed:

# PulseAudio (most desktop distros)
pulseaudio --version

# PipeWire
pw-play --version

# ALSA (fallback)
aplay --version

macOS

cd /path/to/stttype
chmod +x install.sh
./install.sh

If sounddevice fails, install PortAudio:

brew install portaudio

Note: On macOS, you need to grant Accessibility permissions for pynput to capture global hotkeys. Go to System Settings > Privacy & Security > Accessibility and add your terminal application.

Commands

Once installed, stttype works from any terminal.

Command	Description
stttype --start	Start STT Type in background
stttype --shutdown	Stop all STT Type processes
stttype --status	Check if STT Type is running
stttype --restart	Restart STT Type
stttype --config	Open settings GUI
stttype --addtostartup	Add to startup (Windows only)
stttype --rmtostartup	Remove from startup (Windows only)
stttype --model <size>	Set Whisper model (tiny/base/small/medium/large-v3)
stttype --lang <code>	Set language (en/zh/auto/etc)
stttype --help	Show help

Examples

# Start with default settings
stttype --start

# Start with a larger model for better accuracy
stttype --start --model small

# Start with Chinese language
stttype --start --lang zh

# Open settings GUI
stttype --config

Settings GUI

Run stttype --config to open a settings window where you can configure:

Setting	Options
Mode	Streaming (real-time), Hold (legacy)
Whisper Model	tiny, base, small, medium, large-v3
Language	Auto-detect, English, Chinese, Spanish, French, German, Japanese, Korean, Russian, Italian, Portuguese, Arabic, Hindi
Device	Auto (GPU if available), GPU (CUDA), CPU
Compute Type	Auto, Float16 (GPU), Int8 (CPU)
Hotkey	F1 - F12

Settings are saved to ~/.config/stttype/config.json and persist across restarts.

How It Works

Streaming Mode (default)

1. Hold F2 — A transparent red dot appears, microphone starts streaming to local WhisperLive server
2. Speak — Interim transcription appears in the console in real-time
3. Release F2 — Red dot disappears, streaming stops, final text is typed at your cursor

Hold Mode (legacy)

1. Hold F2 — A transparent red dot appears, microphone starts recording
2. Release F2 — Red dot disappears, recording stops
3. Transcribe — Whisper processes the recorded audio
4. Text is pasted — Result is pasted at your cursor position via Ctrl+V

Models

Model	Size	VRAM	Speed	Accuracy
tiny	39 MB	~1 GB	Fastest	Basic
base	74 MB	~1 GB	Fast	Good
small	244 MB	~2 GB	Medium	Better
medium	769 MB	~5 GB	Slower	Best
large-v3	1550 MB	~10 GB	Slowest	Excellent

Default is base - a good balance of speed and accuracy.

Uninstall

Windows

cd "E:\Lucas\STT Type"
.\uninstall.ps1

Linux/macOS

cd /path/to/stttype
chmod +x uninstall.sh
./uninstall.sh

Troubleshooting

Issue	Solution
stttype not found	Restart terminal after installation
"CUDA not available"	Install NVIDIA drivers and CUDA toolkit, or STT Type will auto-fallback to CPU
No sound on start/stop (Linux)	Install pulseaudio-utils, alsa-utils, or pipewire
No overlay on Linux	Ensure $DISPLAY is set (e.g., DISPLAY=:0 stttype --start)
Text not typing	Make sure the target window is focused
Model download fails	Check internet connection
Hotkeys don't work (macOS)	Grant Accessibility permissions to your terminal
Hotkeys don't work (Linux)	Make sure you're running under X11 (not Wayland)
Garbled text when typing	Fixed in v2.0.2+ — uses clipboard paste instead of key simulation
Config not applied	Fixed in v2.0.5+ — CLI defaults changed to None so config values are read
Streaming mode slow on CPU	Use tiny or base model; or switch to Hold mode

Version History

Version	Changes
3.0.0	Major: Real-time streaming via WhisperLive; mode selector (Streaming/Hold); hold-F2 UX preserved
2.0.8	Linux: pyautogui.typewrite() typing; faster bell beeps; transparent PNG overlay via PIL
2.0.7	Linux sound via WAV playback; Linux overlay fixes
2.0.6	Linux sound via WAV playback (paplay/aplay/pw-play); Linux overlay alpha transparency fallback
2.0.5	Fixed config not being used by CLI
2.0.4	README sync
2.0.3	Config value extraction fix
2.0.2	Clipboard typing fix; config applied
2.0.1	Linux DISPLAY fix
2.0.0	Config GUI, transparent indicator
1.0.x	Initial releases

Publish to PyPI

# Install build tools
pip install build twine

# Build
cd /path/to/stttype
python -m build

# Upload
python -m twine upload dist/*

When prompted:

• Username: __token__
• Password: Your PyPI API token

---

Author: LucasApps
Version: 3.0.0
License: MIT