Abstract—In this work, we address the challenging task of single-channel speech separation in realistic, reverberant environments. Our method focuses directly on separating overlapping speech signals captured through a single fixedposition microphone. We collected a custom dataset of crosstalk recordings using a Zoom H5 recorder in two acoustically distinct rooms, involving 50 speakers both male and female engaged in controlled conversational scenarios using standard Harvard sentences that are phonetically rich. Each recording captures dual-speaker overlaps within a single-channel signal, providing a realistic test for deep learning-based separation models. Our approach leverages data-driven neural architectures trained to separate the concurrent speech sources under varied room conditions with the knowledge of room geometry and microphone placement. Experimental results demonstrate the model’s robustness across different spatial configurations and room sizes, showcasing its applicability in real-world speech communication systems. The effectiveness of the separation is evaluated using both objective metrics and perceptual measures, confirming the viability of deploying such systems in practical, resource-constrained settings.

Cite: Jaipreet Kour Wazir and Javaid A. Sheikh, “Speaker Separation in Overlapping Speech Using Single-Channel Recordings in Varied Acoustic Environments," Journal of Communications, vol. 21, no. 1, pp. 139-148, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).