Note
Turn your audio thoughts into notes.
Derecorder is designed to convert voice recorder notes into text. Drag a folder with your recordings into the app and transcribe them locally — no cloud upload required.
Python GUI app for transcribing audio files from a folder using a local OpenAI Whisper model.
Each transcript is formatted as a note with a date and time taken from the recording itself — first from audio metadata (ID3 and other tags), then from the filename, and finally from the file’s creation or modification date if nothing else is available.
When you use Transcribe all, every note is merged into one document and sorted chronologically by date and time, grouped by day.
- Python 3.10+
- ffmpeg — required for mp3, m4a, ogg, and other formats:
# Ubuntu / Debian
sudo apt install ffmpeg python3-tk
# macOS (Homebrew)
brew install ffmpegOn Linux, python3-tk may be needed for drag-and-drop.
Create a virtual environment, install dependencies, and run:
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python -m derecorderOr use the launcher script (creates venv and installs dependencies on first run):
./run.shOn first run, Whisper downloads the selected model weights (e.g. base is ~150 MB).
.mp3, .wav, .flac, .ogg, .m4a, .aac, .opus, .webm, .wma
| Model | Size | Recommendation |
|---|---|---|
| tiny | ~75 MB | Fast, lower quality |
| base | ~150 MB | Balanced (default) |
| small | ~500 MB | Good quality |
| medium | ~1.5 GB | Better with GPU |
| large | ~3 GB | Best quality, GPU recommended |
| turbo | ~1.6 GB | Optimized large-v3 |
On CPU, tiny and base are recommended. medium and large are much faster with CUDA/GPU.