Localized watermarking for AI-generated speech audios, with SOTA on robustness and very fast detector - View it on GitHub