Modified Magnitude-Phase Spectrum Information for Spoofing Detection

Yang, Jian; Wang, Hongji; Das, Rohan Kumar; Ye, Qian

doi:10.1109/taslp.2021.3060810

Cited by 33 publications

(10 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For unknown-kind SSD scenario, Table 3 compares Res2Net-A-C with CQT feature configuration including CQT-MMPS because CQT-MMPS based model showed comparable performance in this scenario [13]. We found that Res2Net-C with CQT+phase outperformed all other methods in LA-Evaluation and -Development, while showing the comparable performance compared to Res2Net-A with CQT in PA-Evaluation and -Development; thus, we claim that Res2Net-C with CQT+phase is appropriate for the practical SSD application.…”

Section: Resultsmentioning

confidence: 99%

“…In [20], the raw waveform is adopted to deploy the phase information with backend neural network named AAIST, showing extraordinary performance in LA, however, there is no demonstration in PA. In [13], CQT with modified magnitude-phase spectrum (MMPS) was proposed to integrate the CQT-magnitude and -phase spectra in a hand-craft way. Although CQT-MMPS outperformed CQT-power spectrum in both PA and LA, there is no guarantee that the CQT-MMPS is an optimal method for combining the CQT-magnitude and -phase spectra for the synergy effect.…”

Section: Motivationmentioning

confidence: 99%

“…We found that a number of researches have investigated various acoustic features to demonstrate their robustness to PA and LA with the state-of-the-art backend neural networks including: ResNet [13][14][15][16][17], Res2Net [18,19], and graph attention network [20,21]. The representative acoustic features for SSD are: log-power (or magnitude) discrete Fourier transform (DFT), constant-Q transform (CQT), and linear frequency cepstral coefficients (LFCC), which are the magnitude features in the frequency domain [16,18,[22][23][24].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Phase-Aware Spoof Speech Detection Based on Res2Net with Phase Network

Kim¹,

Ban²

2022

Preprint

View full text Add to dashboard Cite

The spoof speech detection (SSD) is the essential countermeasure for automatic speaker verification systems. Although SSD with magnitude features in the frequency domain has shown promising results, the phase information also can be important to capture the artefacts of certain types of spoofing attacks. Thus, both magnitude and phase features must be considered to ensure the generalization ability to diverse types of spoofing attacks. In this paper, we investigate the failure reason of feature-level fusion of the previous works through the entropy analysis from which we found that the randomness difference between magnitude and phase features is large, which can interrupt the feature-level fusion via backend neural network; thus, we propose a phase network to reduce that difference. Our SSD system: phase network equipped Res2Net achieved significant performance improvement, specifically in the spoofing attack for which the phase information is considered to be important. Also, we demonstrate our SSD system in both knownand unknown-kind SSD scenarios for practical applications.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Motivationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Phase-Aware Spoof Speech Detection Based on Res2Net with Phase Network

Kim¹,

Ban²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Results for an ablation study show that spectral attention is more important than temporal attention, but that both are beneficial, and that graph pooling improves performance substantially. Last, a comparison of dif- [63] DASC-CQT 0.0940 3.13 LCNN+CE [64] CQT 0.1020 4.07 LCNN [65] LFCC 0.1000 5.06 ResNet [66] CQT-MMPS 0.1190 3.72 ferent systems, working either on the raw signal or hand-crafted inputs show the benefit of a fully end-to-end approach.…”

Section: Discussionmentioning

confidence: 99%

End-to-End Spectro-Temporal Graph Attention Networks for Speaker Verification Anti-Spoofing and Speech Deepfake Detection

Tak¹,

Jung²,

Patino³

et al. 2021

Preprint

View full text Add to dashboard Cite

Artefacts that serve to distinguish bona fide speech from spoofed or deepfake speech are known to reside in specific subbands and temporal segments. Various approaches can be used to capture and model such artefacts, however, none works well across a spectrum of diverse spoofing attacks. Reliable detection then often depends upon the fusion of multiple detection systems, each tuned to detect different forms of attack. In this paper we show that better performance can be achieved when the fusion is performed within the model itself and when the representation is learned automatically from raw waveform inputs. The principal contribution is a spectro-temporal graph attention network (GAT) which learns the relationship between cues spanning different sub-bands and temporal intervals. Using a model-level graph fusion of spectral (S) and temporal (T) sub-graphs and a graph pooling strategy to improve discrimination, the proposed RawGAT-ST model achieves an equal error rate of 1.06% for the ASVspoof 2019 logical access database. This is one of the best results reported to date and is reproducible using an open source implementation.

show abstract

“…The features extracted are trained using machine learning algorithms ranging from generative models like i-vectors [15], Gaussian Mixture Models (GMM) [10], [16], Universal Background Models (UBM) [17], [18], [19] and Joint Factor Analysis [15] to discriminative models like Support Vector Machines (SVM) [20], Deep Neural Networks [21], [22], [23] and its variants like Recurrent Neural Networks (RNN) [24], [25], Deep Residual Neural Networks [13] and Convolutional Neural Network (CNN) [26], [27]. The GMM are considered to be efficient in capturing the generality and non-linearities in data [2].…”

Section: Related Workmentioning

confidence: 99%

A Unique Glottal Flow Parameters based Features for Anti-spoofing Countermeasures in Automatic Speaker Verification

Chadha¹,

Abdullah²,

Angeline³

2021

IJACSA

View full text Add to dashboard Cite

The domain of Automatic Speaker Verification (ASV) is blooming with growing developments in feature engineering and artificial intelligence. Inspite of this, the system is liable to spoofing attacks in the form of synthetic or replayed speech. The difficulty in detecting synthetic speech is due to recent advancements in the Voice conversion and Text-to-speech systems which produce natural, indistinguishable speech. To prevent such attacks, there is a need to develop robust spoof detection systems. In order to achieve this goal, we are proposing estimation of Glottal Flow Parameters (GFP) from speech of genuine speech and synthetic spoof samples. The GFP are further parameterized using time, frequency and Liljencrants-Fant (LF) models. Along with GFP features, the Linear Prediction Cepstrum Co-efficient (LFCC) and statistical parameters are computed. The GFP features are investigated to prove their usefulness in detecting spoofed and genuine speech. The ASV spoof 2019 corpus is used to test the framework and evaluated against the baseline models. The proposed spoof detection framework produces an Equal Error Rate (EER) of 2.39% and tandem Detection Cost Function (t-DCF) of 0.0562 which is found to be better than the state-of-the art technique.

show abstract

Modified Magnitude-Phase Spectrum Information for Spoofing Detection

Cited by 33 publications

References 52 publications

Phase-Aware Spoof Speech Detection Based on Res2Net with Phase Network

Phase-Aware Spoof Speech Detection Based on Res2Net with Phase Network

End-to-End Spectro-Temporal Graph Attention Networks for Speaker Verification Anti-Spoofing and Speech Deepfake Detection

A Unique Glottal Flow Parameters based Features for Anti-spoofing Countermeasures in Automatic Speaker Verification

Contact Info

Product

Resources

About