On the Impact of FFP2 Face Masks on Speaker Verification for Mobile Device Authentication

David Sedlak, Rainhard Dieter Findling

Research output: Chapter in Book/Report/Conference proceedingsConference contributionpeer-review

1 Downloads (Pure)

Abstract

Voice-based authentication can allow for straightforward and unobtrusive authentication with mobile devices. With COVID-19, wearing face masks has become common in many parts of the world. While prior research has shown that face masks act like low pass filters, the impact of face masks on voice-based authentication with mobile devices is still unclear. In this paper we analyze the impact of FFP2 masks on voice authentication for mobile devices in realistic scenarios. We evaluate a pretrained EPACA-TDDN speaker verification model with a self-recorded database of 450 mobile voice authentication utterances from 30 speakers, with a total length of 29 min. Results indicate that wearing FFP2 face masks has a slight but significant impact on speaker verification scores. Results also indicate that in authentication scenarios they lead to a slight increase of the FNR, and in comparison smaller decrease of the FPR, and in direct comparison, to a slight increase of the EER.
Original languageEnglish
Title of host publicationAdvances in Mobile Computing and Multimedia Intelligence - 21st International Conference, MoMM 2023, Proceedings
EditorsPari Delir Haghighi, Eric Pardede, Gillian Dobbie, Vithya Yogarajan, Ngurah Agus Sanjaya ER, Gabriele Kotsis, Ismail Khalil
Pages35-49
Number of pages15
DOIs
Publication statusPublished - 2023

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14417 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Keywords

  • speaker verification
  • mobile authentication
  • face masks

Fingerprint

Dive into the research topics of 'On the Impact of FFP2 Face Masks on Speaker Verification for Mobile Device Authentication'. Together they form a unique fingerprint.

Cite this