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

David Sedlak, Rainhard Dieter Findling

Publikation: Beitrag in Buch/Bericht/TagungsbandKonferenzbeitragBegutachtung

10 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.
OriginalspracheEnglisch
TitelAdvances in Mobile Computing and Multimedia Intelligence - 21st International Conference, MoMM 2023, Proceedings
Redakteure/-innenPari Delir Haghighi, Eric Pardede, Gillian Dobbie, Vithya Yogarajan, Ngurah Agus Sanjaya ER, Gabriele Kotsis, Ismail Khalil
Seiten35-49
Seitenumfang15
DOIs
PublikationsstatusVeröffentlicht - 2023

Publikationsreihe

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Band14417 LNCS
ISSN (Print)0302-9743
ISSN (elektronisch)1611-3349

Fingerprint

Untersuchen Sie die Forschungsthemen von „On the Impact of FFP2 Face Masks on Speaker Verification for Mobile Device Authentication“. Zusammen bilden sie einen einzigartigen Fingerprint.

Zitieren