Real Time Voice Isolation and Identification in Multi-speaker Environments

Real-time processing of audio containing multiple overlapping speakers remains a challenging task, particularly in everyday environments such as classrooms, meetings, and group discussions.Most existing speechprocessing systems perform well only under controlled conditions and often fail to maintain speaker separation and clarity when overlap and background noise are present.This paper presents a system designed to isolate individual speakers and generate structured, speaker-wise transcripts from multi-speaker audio recordings.The proposed pipeline combines audio validation, preprocessing, speaker diarization, and segmentwise automatic speech recognition to handle overlapping speech more effectively.Speaker segmentation is performed prior to transcription to reduce speaker confusion and improve transcript consistency.The system is implemented using a modular, cloud-supported architecture that allows parallel processing and scalable deployment.Experimental observations across recordings with varying noise levels indicate that the system maintains stable speaker attribution and produces readable transcripts in practical scenarios.The proposed approach aims to support real-world applications where accurate multi-speaker transcription is required without reliance on specialized recording hardware.