Graduate School and Research Center in Digital Sciences

Integrated presentation attack detection and automatic speaker verification: Common features and Gaussian back-end fusion

Todisco, Massimiliano; Delgado, Héctor; Lee, Kong Aik; Sahidullah, Md; Evans, Nicholas; Kinnunen, Tomi; Yamagishi, Junichi

INTERSPEECH 2018, 19th Annual Conference of the International Speech Communication Association, September 2-6, 2018, Hyderabad, India

The vulnerability of automatic speaker verification (ASV) systems to spoofing is widely acknowledged. Recent years have seen an intensification in research efforts to develop spoofing countermeasures, also known as presentation attack detection (PAD) systems. Much of this work has involved the exploration of features that discriminate reliably between bona fide and spoofed speech. While there are grounds to use different frontends for ASV and PAD systems (they are different tasks) the use of a single front-end has obvious benefits, not least convenience and computational efficiency, especially when ASV and PAD are combined. This paper investigates the performance of a variety of different features used previously for both ASV and PAD and assesses their performance when combined for both tasks. The paper also presents a Gaussian back-end fusion approach to system combination. In contrast to cascaded architectures, it relies upon the modelling of the two-dimensional score distribution stemming from the combination of ASV and PAD in parallel. This approach to combination is shown to generalise particularly well across independent ASVspoof 2017 v2.0 development and evaluation datasets.

Document Doi Hal Bibtex

Title:Integrated presentation attack detection and automatic speaker verification: Common features and Gaussian back-end fusion
Keywords:automatic speaker verification, spoofing, countermeasures, presentation attack detection
Type:Conference
Language:English
City:Hyderabad
Country:INDIA
Date:
Department:Digital Security
Eurecom ref:5573
Copyright: © ISCA. Personal use of this material is permitted. The definitive version of this paper was published in INTERSPEECH 2018, 19th Annual Conference of the International Speech Communication Association, September 2-6, 2018, Hyderabad, India and is available at : http://dx.doi.org/10.21437/Interspeech.2018-2289
Bibtex: @inproceedings{EURECOM+5573, doi = {http://dx.doi.org/10.21437/Interspeech.2018-2289}, year = {2018}, title = {{I}ntegrated presentation attack detection and automatic speaker verification: {C}ommon features and {G}aussian back-end fusion}, author = {{T}odisco, {M}assimiliano and {D}elgado, {H}{\'e}ctor and {L}ee, {K}ong {A}ik and {S}ahidullah, {M}d and {E}vans, {N}icholas and {K}innunen, {T}omi and {Y}amagishi, {J}unichi}, booktitle = {{INTERSPEECH} 2018, 19th {A}nnual {C}onference of the {I}nternational {S}peech {C}ommunication {A}ssociation, {S}eptember 2-6, 2018, {H}yderabad, {I}ndia}, address = {{H}yderabad, {INDIA}}, month = {09}, url = {http://www.eurecom.fr/publication/5573} }
See also: