A Study of Using Cepstrogram for Countermeasure Against Replay Attacks

Authors: Shih-Kuang Lee, Yu Tsao, Hsin-Min Wang

Published: 2022-04-09 00:18:53+00:00

Comment: Submitted to SLT 2022

AI Summary

This study investigates the cepstrogram properties and demonstrates its effectiveness as a powerful countermeasure against replay attacks in automatic speaker verification (ASV) systems. A cepstrum analysis suggests that crucial anti-spoofing information for replay attacks is retained in the cepstrogram. Experiments show that cepstrogram-based single and fusion systems, particularly with an LCNN backend, significantly outperform existing state-of-the-art methods on the ASVspoof 2019 physical access database.

Abstract

This study investigated the cepstrogram properties and demonstrated their effectiveness as powerful countermeasures against replay attacks. A cepstrum analysis of replay attacks suggests that crucial information for anti-spoofing against replay attacks may be retained in the cepstrogram. When building countermeasures against replay attacks, experiments on the ASVspoof 2019 physical access database demonstrate that the cepstrogram is more effective than other features in both single and fusion systems. Our LCNN-based single and fusion systems with the cepstrogram feature outperformed the corresponding LCNN-based systems without the cepstrogram feature and several state-of-the-art single and fusion systems in the literature.

Key findings

The cepstrogram feature proved more effective than other conventional features (CQT, LFCC, DCT, Spectrogram) in single systems for detecting replay attacks. Fusion systems incorporating the cepstrogram consistently achieved better performance, with the LCNN-CQT+LFCC+Spec+Ceps fusion system achieving a new state-of-the-art EER of 0.094 on the ASVspoof 2019 PA evaluation set, significantly outperforming previous methods.

Approach

The authors propose using the cepstrogram as a discriminative front-end feature for anti-spoofing against replay attacks. They use a Light Convolutional Neural Network (LCNN) architecture as the backend classifier, evaluating its performance with cepstrogram alone and in fusion systems with other features like CQTgram, LFCC, and spectrogram.

Datasets

ASVspoof 2019 physical access (PA) database

Model(s)

Light Convolutional Neural Network (LCNN)

Author countries

Taiwan

← Previous