Conversations stored as mono data is a common problem in many real world speaker recognition applications. In this paper, we focus on investigative scenarios, where a number of mono telephone conversations are available for a speaker of interest. For example, a human operator may have verified that the speaker is present in these conversations. We propose several approaches for automatically creating enrollment models for the speaker of interest from such data. We then use the enrollment models to search for appearances of the speaker of interest in other calls. We analyze the performance of the different method on two dataset that matches our scenario, one is from a simulated case and one is from a real case.
Cite as: Solewicz, Y., Cohen, N., Rohdin, J., Madikeri, S., Čercnocký, J.”. (2022) Speaker Recognition on Mono-Channel Telephony Recordings. Proc. The Speaker and Language Recognition Workshop (Odyssey 2022), 193-199, doi: 10.21437/Odyssey.2022-27
@inproceedings{solewicz22_odyssey,
author={Yosef Solewicz and Noa Cohen and Johan Rohdin and Srikanth Madikeri and Jan ”Honza” Čercnocký},
title={{Speaker Recognition on Mono-Channel Telephony Recordings}},
year=2022,
booktitle={Proc. The Speaker and Language Recognition Workshop (Odyssey 2022)},
pages={193--199},
doi={10.21437/Odyssey.2022-27}
}