ghe_transcribe

A tool to transcribe audio files with speaker diarization using Faster Whisper and Pyannote.

• Fast transcription with optimized Whisper models
• Speaker diarization to identify different speakers
• Multiple output formats (TXT, SRT)
• Jupyter interface for interactive use
• CLI tool for global compatibility

Interface Preview

The Jupyter-based interface provides an intuitive way to upload audio files, configure transcription settings, and download results in multiple formats.

Installation

System Dependencies

This tool requires FFmpeg for audio processing:

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt update && sudo apt install ffmpeg

# Windows
choco install ffmpeg

ghe_transcribe

git clone https://github.com/Global-Health-Engineering/ghe_transcribe.git
cd ghe_transcribe
python -m venv venv
source venv/bin/activate
pip install -e .

Note

See the detailed installation guide.

Hugging Face Authentication

This tool uses gated models from Hugging Face that require authentication. You need to:

1. Join Hugging Face, to access Pyannote
   • https://hf.co/join
2. Accept User Conditions, to use Pyannote
   • https://hf.co/pyannote/speaker-diarization-3.1
   • https://hf.co/pyannote/segmentation-3.0
3. Create Access Token, to use ghe_transcribe
   • https://hf.co/settings/tokens

Usage

Renkulab

See the detailed documentation for Renkulab

Jupyter Interface (Local)

Open app.ipynb and run the cell:

from ghe_transcribe.app import execute
execute()

Python API

from ghe_transcribe.core import transcribe
result = transcribe("media/test01.mp3")

Command Line

# Simplest call
transcribe media/test01.mp3

# Multiple files
transcribe media/test01.mp3 media/test02.m4a --trim 5

# See all options
transcribe --help 

Editors

• For SRT files subtitle-editor.org/, runs locally on your browser
• For TXT files note-taking apps, Word, MAXQDA, QualCoder, ...

Contributing

We welcome contributions! Please use conventional commits. See our contributions guidelines.

License

MIT License - see LICENSE.