OpenWhisper

Free, open-source voice-to-text for macOS. Hold a key, speak, release — text appears wherever your cursor is. Works in any app.

Like Wispr Flow but 100% free, fully local, and private. No cloud, no subscription, no data leaves your Mac.

How It Works

1. OpenWhisper runs in your menu bar (🎙)
2. Hold Right Option → speak → release
3. OpenAI Whisper transcribes your speech locally
4. Ollama + Qwen 2.5 cleans up filler words and grammar
5. Text is auto-pasted at your cursor — VS Code, Terminal, Slack, browser, anywhere

Fully local. Fully private. Zero cost.

Demo

Hold Right ⌥ → "Hey let's um add a new function that uh parses the JSON response"
Output:      → "Hey, let's add a new function that parses the JSON response."

Install

Quick Install (one command)

curl -sSL https://raw.githubusercontent.com/Rajvardhman05/openwhisper/main/install.sh | bash

Manual Install

# 1. Install system deps
brew install ffmpeg ollama

# 2. Install Python packages
pip3 install openai-whisper sounddevice numpy pynput rumps

# 3. Start Ollama and pull cleanup model
brew services start ollama
ollama pull qwen2.5:3b

# 4. Clone and run
git clone https://github.com/Rajvardhman05/openwhisper.git ~/openwhisper
python3 ~/openwhisper/openwhisper_app.py

macOS Permissions (required)

After first launch, grant these in System Settings → Privacy & Security:

Permission	What to add
Accessibility	python3 (find it with which python3)
Microphone	Terminal / iTerm (whichever you use)

Usage

Menu Bar App (recommended)

python3 openwhisper_app.py

Runs as a menu bar icon. Hold Right Option to dictate.

Terminal Mode

python3 openwhisper.py           # Hold Right Option to talk
python3 openwhisper.py -c        # Continuous mode
python3 openwhisper.py --raw     # Skip LLM cleanup
python3 openwhisper.py -m small  # Better accuracy (larger model)

Options

Flag	Description
-m, --model	Whisper model: tiny, base, base.en, small, medium, large
-l, --language	Language code (default: en)
-k, --key	Trigger key: option_r, ctrl_r, cmd_r, shift_r
--raw	Skip LLM cleanup

Environment Variables

OPENWHISPER_MODEL=small python3 openwhisper_app.py   # Use larger model
OPENWHISPER_LANG=es python3 openwhisper_app.py       # Spanish

Architecture

Microphone → Whisper (local STT) → Ollama/Qwen (cleanup) → CGEvent (paste)

Component	Purpose	Size
OpenAI Whisper	Speech-to-text	~140MB (base)
Ollama + Qwen 2.5 3B	Filler removal, grammar	~1.9GB
CGEvent (macOS)	Auto-paste at cursor	Built-in
rumps	Menu bar UI	Tiny
pynput	Global hotkey	Tiny

Total disk: ~2.5GB (models cached locally)

Whisper Models

Model	Size	Speed	Accuracy	Best for
tiny	39MB	Fastest	Basic	Quick notes
base	140MB	Fast	Good	Daily use
base.en	140MB	Fast	Better (English)	Recommended
small	460MB	Medium	Great	Important dictation
medium	1.5GB	Slow	Excellent	Maximum accuracy

Roadmap

• Core dictation (hold-to-talk, auto-paste)
• LLM cleanup (filler removal, grammar)
• Menu bar app
• Native macOS app (Swift/SwiftUI)
• Custom app icon
• Homebrew cask (brew install openwhisper)
• Auto-start at login
• Configurable hotkey via UI
• Snippet support (voice shortcuts)
• iOS / iPadOS app
• Context-aware tone (formal in email, casual in chat)

Why OpenWhisper?

Feature	Wispr Flow	SuperWhisper	OpenWhisper
Price	$15/mo	$8.50/mo	Free forever
Privacy	Cloud-based	Local	Fully local
Offline	No	Yes	Yes
Filler removal	Yes	No	Yes (LLM)
Open source	No	No	Yes
Works in any app	Yes	Yes	Yes

Contributing

PRs welcome! See the roadmap above for what's needed.

License

MIT — use it however you want.

Support

If OpenWhisper saves you time, consider supporting development: