Idiap study explores impact of score fusion on voice biometrics, presentation attack detection

May 18, 2017 - 

Researchers at Idiap Research Institute Biometrics Group in Switzerland have conducted an extensive study of eight presentation attack detection (PAD methods) in which they assessed their ability to detect known and unknown attacks using publically available speaker databases with spoofing attacks, AVspoof and ASVspoof.

Authored by research associate Pavel Korshunov and senior researcher Sébastien Marcel, the study is published in the latest issue of IEEE Journal of Selected Topics in Signal Processing.

Automatic speaker verification (ASV) systems are highly vulnerable to spoofing or presentation attacks, preventing their wide deployment.

To rectify this, the researchers sought to develop mechanisms that can detect these types of attacks, as well as be seamlessly integrated into existing ASV systems for practical and attack-resistant solutions.

In order for them to be practical solutions, an attack detection should have high accuracy, be well-generalized for different attacks, and be simple and efficient.

And although there have been many audio-based presentation attack detection (PAD) methods proposed as of late, their evaluation was typically conducted on a single, often obscure, database with limited number of attacks.

Korshunov and Marcel decided to investigate whether combining multiple PAD systems via score fusion can help improve attack detection accuracy.

Score fusion refers to methods in which several biometric samples, instances, or modalities are compared, and the resulting probabilities are combined to form a single fused score.

In addition, the researchers explored the effects of combining PAD systems (via parallel and cascading schemes) with two i-vector and inter-session variability based ASV systems on the overall performance in both bona fide scenarios (no attacks) and in scenarios where spoofing occurred.

The test results question the efficiency and practicality of the existing PAD systems, particularly when comparing results for individual databases and cross-database data.

Fusing many PAD systems can result in a slightly improved performance, but the researchers are still unsure as to which systems they should select for combined use.

Joint ASV-PAD systems displayed a significantly increased resistance to the attacks but yielded a slightly lower performance for bona fide scenarios.

Leave a Comment

comments

About Justin Lee

Justin Lee has been a contributor with Biometric Update since 2014. Previously, he was a staff writer for web hosting magazine and website, theWHIR. For more than a decade, Justin has written for various publications on issues relating to technology, arts and culture, and entertainment. Follow him on Twitter @BiometricJustin.