wq2012/awesome-diarization

Organizes diarization research across modular subsections—supervised and online approaches, speaker embedding techniques, clustering methods, and emerging LLM-based post-processing—alongside standardized benchmark datasets (DIHARD, AISHELL-4) and evaluation tools. Covers the full diarization pipeline including speaker change detection, audio feature extraction, and data augmentation strategies, plus audio-visual extensions and joint ASR-diarization systems. Serves as a reference for both classical clustering-based and modern end-to-end neural architectures handling variable speaker counts and overlapped speech scenarios.

1,851 stars. No commits in the last 6 months.