whisper-drop

A minimal macOS app to transcribe audio files with drag & drop, powered by whisper.cpp and quantized GGML models.

---

Features

• Drag & drop any audio or video file onto the window
• Supports .mp3, .wav, .m4a, .ogg, .flac, .opus, .webm, .mp4, .aac
• Quantized GGML models (Q5/Q8) for lower resource usage on CPU
• Models are downloaded automatically on first use and cached locally
• Two-column layout with live log panel
• Outputs a .txt file saved next to the original audio
• No technical knowledge required to use

---

Requirements

• macOS (Apple Silicon or Intel)
• Internet connection for the first setup and first use of each model

---

Installation

Clone the repo:

git clone https://github.com/your-username/whisper-drop.git
cd whisper-drop
chmod +x WhisperDrop.command
chmod +x WhisperDrop_installer.command

Then run the installer:

./scripts/setup.sh

Or double-click scripts/WhisperDrop_installer.command from Finder.

This will automatically install:

• Homebrew
• ffmpeg
• whisper.cpp (whisper-cli)
• Python 3.11
• tkinterdnd2 (in a local .venv)

Note: Setup takes a few minutes the first time.

---

Usage

Double-click WhisperDrop.command to open the app.

On first launch, WhisperDrop.command will run setup automatically if it hasn't been done yet.

Gatekeeper warning: macOS may block .command files downloaded from the internet. If you see a "Not Opened" warning, click Done, then run this once in Terminal:

xattr -d com.apple.quarantine /path/to/WhisperDrop.command
xattr -d com.apple.quarantine /path/to/scripts/setup.sh
xattr -d com.apple.quarantine "/path/to/scripts/WhisperDrop_installer.command"

Once the app is open:

1. Drag & drop an audio file onto the window (or click to browse)
2. Select the language and model
3. Click Start Transcription
4. The .txt file is saved in the same folder as the original audio

---

Models

Models are downloaded from HuggingFace on first use and cached in .models/whisper.cpp/ inside the app folder.

Model	Size	Speed	Best for
Tiny Q5	~32 MB	Fastest	Older or slower hardware
Base Q5	~57 MB	Fast	Balanced choice for most users
Small Q5	~190 MB	Medium	Better accuracy on complex audio
Medium Q5	~515 MB	Slow	High accuracy
Turbo Q5	~547 MB	Fast	Fast and accurate, recommended
Turbo Q8	~874 MB	Medium	Maximum accuracy while quantized

Turbo Q5 is the default and recommended for most use cases.

---

Development

To run the app with hot reload while editing transcriber.py:

# Install watchdog once
.venv/bin/pip install watchdog

# Start the dev watcher
.venv/bin/python scripts/dev.py

Every time you save transcriber.py the app will restart automatically.

---

Project structure

whisper-drop/
├── scripts/
│   ├── setup.sh                       # Full setup script
│   └── dev.py                         # Hot-reload dev launcher
├── transcriber.py                     # Main app
├── WhisperDrop.command                # Launcher (double-click to open)
├── WhisperDrop_installer.command  # Explicit installer
├── requirements.txt                   # Python dependencies
├── .gitignore
├── LICENSE
├── README.md
├── .venv/                             # Local Python environment (created by setup)
└── .models/whisper.cpp/               # Cached GGML models (created on first use)

---

License

MIT