Speaker diarization—the task of partitioning an audio stream into speaker-specific segments—has become essential for analyzing human interactions in real-world settings. Recent breakthroughs in deep learning and large language models (LLMs) have dramatically improved both accuracy and processing speed. This paper presents an in-depth review of AI-powered speaker diarization methods, detailing system architectures, numerical performance metrics, and ethical considerations. Experimental results demonstrate significant improvements in Diarization Error Rates (DER), with our system reducing errors by up to 40% compared to traditional approaches. The applications range from enterprise communications to educational analytics, offering promising advancements for next-generation conversational AI.