./Hush

Overview

Hush is a local speech-to-text tool that runs entirely on your machine. Press a hotkey to start recording, speak, and when you stop the transcription is automatically copied to your clipboard and pasted. No cloud APIs, no subscriptions -- just OpenAI's Whisper model running locally.

Features

• Hotkey-driven -- Shift+F10 by default, fully customizable via keymaps.txt
• Auto-paste -- transcription is copied to clipboard and pasted automatically after recording stops
• Multiple models -- choose from tiny (75MB, fastest) to large (3GB, best accuracy)
• Multi-language -- supports transcription and translation (e.g., Russian to English)
• Standalone binary -- builds to a single hush.exe via PyInstaller
• Fully local -- no internet required after model download

Usage

hush.exe          # Default model (base)
hush.exe tiny     # Fastest, lower accuracy
hush.exe small    # Good accuracy, recommended for non-English
hush.exe medium   # Great accuracy, slower
hush.exe large    # Best accuracy, slowest

How It Works

1. Hush loads the selected Whisper model on startup
2. Press the hotkey to start recording audio via sounddevice
3. Press the hotkey again to stop
4. Audio is transcribed locally through Whisper
5. Result is copied to clipboard via pyperclip and pasted via pynput

Models

| Model | Size | Speed | Accuracy | |-------|------|-------|----------| | tiny | 75MB | Fastest | Basic | | base | 150MB | Fast | OK | | small | 500MB | Medium | Good | | medium | 1.5GB | Slow | Great | | large | 3GB | Slowest | Best |