Dictate

Local dictation app for Windows, macOS-style — press a hotkey, speak, get text pasted at cursor. Powered by faster-whisper.

Features

• NumPad Insert to toggle recording (regular Insert ignored)
• Auto-stop after 30s of silence or 5 minutes total
• System tray — icon with right-click menu, no console window
• CUDA GPU acceleration (int8, ~5s model load time)
• VAD + beam search for punctuation and sentence boundaries

Requirements

• Python 3.11+
• NVIDIA GPU with drivers
• Windows 10/11

Quick Start

pip install -r requirements.txt
python dictate.py

The model (~1.5 GB) downloads automatically on first run from HuggingFace.

Options

python dictate.py --no-tray          # keep console open (debug mode)
python dictate.py --model small      # smaller/faster model
python dictate.py --device "Name"    # select microphone
python dictate.py --language en      # language (default: ru)

Standalone .exe

# Set up clean venv first
uv python install 3.12
uv venv --python 3.12 .venv
.venv\Scripts\pip install -r requirements.txt pyinstaller

# Copy nvidia DLLs into vendor/ (required, not in git)
robocopy .venv\Lib\site-packages\nvidia\cublas\bin vendor\nvidia\cublas\bin cublas64_12.dll
robocopy .venv\Lib\site-packages\nvidia\cuda_nvrtc\bin vendor\nvidia\cuda_nvrtc\bin nvrtc64_120_0.dll nvrtc-builtins64_129.dll

# Build
.venv\Scripts\pyinstaller Dictate.spec
# Output: dist\Dictate.exe (~630 MB)

Files

File	Purpose
dictate.py	Main application
Dictate.spec	PyInstaller config
requirements.txt	Python dependencies
vendor/	NVIDIA DLLs for .exe build (not in repo)