AI-Driven Transcription & Diarization Tool for Twitch VODs

This project provides GPU-accelerated transcription and speaker diarization for audio/video files, with a focus on processing Twitch VODs.

Requirements

• Python 3.9+
• NVIDIA GPU with CUDA support (optional but recommended)
• FFmpeg (for pydub)

Installation

pip install -r requirements.txt

Usage

python main.py <input_file.mp4> -m base -o transcript.json

Modules

• transcribe.py: Uses OpenAI Whisper for transcription.
• diarize.py: Uses pyannote.audio for speaker diarization.
• silence_detect.py: Uses pydub to detect silences.
• main.py: Orchestrates the pipeline and outputs JSON.

Output Format

Each segment in the output JSON includes:

• start: float (seconds) segmentation start time
• end: float (seconds) segmentation end time
• text: string transcript text
• speaker: speaker label (e.g., 'Speaker 1')

Testing

Run unit tests with:

pytest