This project provides a Gradio-based web UI for the Whisper speech-to-text model, enabling easy subtitle generation from various sources like files, YouTube, and microphones. It targets users needing efficient and versatile subtitle creation, offering features like speech-to-text translation and subtitle file translation.

How It Works

The UI integrates multiple Whisper implementations, defaulting to SYSTRAN/faster-whisper for optimized VRAM usage and speed. It supports pre-processing audio with Silero VAD, BGM separation with UVR, and post-processing with pyannote for speaker diarization. Translation capabilities are extended via Facebook NLLB models and the DeepL API.

Quick Start & Requirements

• Install: Clone the repository and run install.bat or install.sh to set up dependencies in a virtual environment. Run start-webui.bat or start-webui.sh to launch.
• Prerequisites: Git, Python (3.10-3.12), FFmpeg (added to PATH), and CUDA (defaulting to 12.4, requires manual adjustment for other versions/hardware).
• Docker: Build image (docker compose build) and run (docker compose up).
• Resources: Image size is ~7GB. Faster-whisper significantly reduces VRAM usage compared to the original OpenAI implementation.
• Docs: Wiki

Highlighted Details

• Supports OpenAI/Whisper, SYSTRAN/faster-whisper, and insanely-fast-whisper.
• Generates SRT, WebVTT, and TXT subtitle formats.
• Includes speech-to-text translation to English and subtitle file translation.
• Integrates speaker diarization via pyannote (requires Huggingface token and term acceptance).
• Offers BGM separation using UVR.

Maintenance & Community

• Active development with several features marked as completed in the TODO list.
• Community contributions for translations are welcomed.

Licensing & Compatibility

• The repository does not explicitly state a license in the provided README.

Limitations & Caveats

• Requires manual configuration of --extra-index-url in requirements.txt for non-Nvidia GPUs or different CUDA versions.
• Pyannote speaker diarization requires manual Huggingface token setup and term acceptance.
• Real-time microphone transcription is listed as a future feature.