Experimental Analysis of Features for Replay Attack Detection — Results on the ASVspoof 2017 Challenge

Abstract

This paper presents an experimental comparison of different features for the detection of replay spoofing attacks in Automatic Speaker Verification systems. We evaluate the proposed countermeasures using two recently introduced databases, including the dataset provided for the ASVspoof 2017 challenge. This challenge provides researchers with a common framework for the evaluation of replay attack detection systems, with a particular focus on the generalization to new, unknown conditions (for instance, replay devices different from those used during system training). Our cross-database experiments show that, although achieving this level of generalization is indeed a challenging task, it is possible to train classifiers that exhibit stable and consistent results across different experiments. The proposed approach for the ASVspoof 2017 challenge consists in the score-level fusion of several base classifiers using logistic regression. These base classifiers are 2-class Gaussian Mixture Models (GMMs) representing genuine and spoofed speech respectively. Our best system achieves an Equal Error Rate of 10.52% on the challenge evaluation set. As a result of this set of experiments, we provide some general conclusions regarding feature extraction for replay attack detection and identify which features show the most promising results.

ASVspoof 2017 Version 2.0: meta-data analysis and baseline enhancements

“…Other researchers have reported the benefit of using CMVN for spoofing detection, e.g. [21,22,23,24], all of which relate to replay detection within the scope of ASVspoof 2017. The application of CMVN to replay spoofing detection may at first seem counter-intuitive.…”

Speech Replay Detection with x-Vector Attack Embeddings and Spectral Features

“…The SCMC feature captures the magnitude of energy a sub-band, which can effectively distinguish two signals even if they share the same average energy. The SCMC feature was also one of the best-performing features from the analysis in [2] for the ASVspoof 2017 challenge, though was based on different data. It has been recognized as a stable feature across experimental conditions.…”

“…Following from the features analysis for replay attack detection that was presented in [2] for the ASVspoof 2017 challenge, we extracted the following features: Mel Frequency Cepstral Coefficients (MFCCs), Inverted Mel Frequency Cepstral Coefficients (IMFCCs), Rectangular Filter Cepstral Coefficients (RFCCs), Linear Frequency Cepstral Coefficients (LFCCs), Sub-band Spectral Centroid Magnitude Coefficients (SCMCs), and Constant Q Cepstral Coefficients (CQCCs) [1]. A description of the features is provided in Table 1.…”

“…Our approach captured these conditions using a convolutional neural network (CNN) architecture, similar to the top system from the earlier ASVspoof 2017 replay detection challenge [3]. We explored several types of signal features such as those from [2] and our final submission was based on our own specially-trained x-vector [5] embeddings combined with signal features.…”

Deep Siamese Architecture Based Replay Detection for Secure Voice Biometric

“…Current countermeasures against replay attacks generally operate in one of three ways: (a) identifying exact reproduction of a previous access attempt; (b) exploiting differences in the speech transmission channel [9]; or (c) targeting artefacts in replayed speech such as pop-noise [10], and source features [11]. Commonly used spectral features include sub-band spectral centroid magnitude coefficients (SCMCs) [12], constant-Q cepstral coefficients (CQCCs) [13], single frequency filtering cepstral coefficients (SFF-CCs) [14], inverse Mel frequency cepstral coefficients (IMFCCs) [15], rectangular filter cepstral coefficients (RFCCs) [15], and scattering decomposition based features [16]. In addition, deep neural network (DNN) architectures have also been employed either as discriminative feature extractors [17] or as an end-toend spoofing detectors [18] in a number of ways.…”