Linear prediction residual features for automatic speaker verification anti-spoofing

Hanilçi, Cemal

doi:10.1007/s11042-017-5181-0

Cited by 8 publications

(10 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These features were first presented in Hanilç (2018). As shown in Figure 3, they can be obtained by applying discrete cosine transform (DCT) to the magnitude of Hilbert transformation computed from the LP residuals.…”

Section: The Features Based On Linear Prediction Analysismentioning

confidence: 99%

“…In Yu, Tan, Ma, Martin, and Guo (2017) authors use different types of filter banks such as linear, Gammatone, and its inverted version and also inverted Mel filter banks to extract different variations of cepstral coefficients for replay attack detection. Furthermore, recently, linear prediction residual‐based features like linear prediction residual magnitude cepstral coefficients (Hanilç, 2018), linear prediction residual phase cepstral coefficients (Hanilç, 2018), LP‐based relative phase features (Phapatanaburi, Wang, Nakagawa, & Iwahashi, 2019), and residual Mel frequency cepstral coefficients (Mishra, Singh, & Pati, 2018), have shown interesting and promising results in detection of spoofed speech.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Evolutionary fusion of classifiers trained on linear prediction based features for replay attack detection

Nasersharif

Yazdani

2021

Expert Systems

View full text Add to dashboard Cite

Recently, linear prediction analysis (LP) related features have been successfully used for replay attack detection due to the imperfection in the LP‐based signal produced by recording and playback devices. In this paper, we propose a weighted linear combination of classifier scores for replay attack detection where our classifiers, including Gaussian mixture models (GMMs) and support vector machines (SVMs), are trained on a variety of LP and LP residual‐based features. In this way, we can benefits from all of the LP‐related features when we combine classifiers trained on these features. We determine classifier weights using two evolutionary algorithms: genetic algorithm and particle swarm optimization. Furthermore, we propose a new feature based on performing LP residuals analysis of Mel sub‐band energies. We also propose a deep structure for extracting deep features from LP‐based coefficients to consider the class labels (genuine or spoofed speaker) in the feature extraction process. Results of our classifier system on the ASVspoof 2017 version 2 dataset show equal error rates of 0.3% and 4.8% for its development and evaluation subset, respectively. We also applied our proposed replay attack detection method to an ASV system that has acceptable results.

show abstract

Section: The Features Based On Linear Prediction Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Evolutionary fusion of classifiers trained on linear prediction based features for replay attack detection

Nasersharif

Yazdani

2021

Expert Systems

View full text Add to dashboard Cite

show abstract

“…Linear prediction residual (LPR) signals that are computed using linear prediction analysis (LPA) [22], has attracted attention in spoofing attack detection [23]- [25] because the LPR is highly affected by the distortion of spoofing process; thus, the difference between original speech and spoofing signal can be found and is anticipated to be useful for spoofing attack countermeasures. In [24], LPR Hilbert envelope cepstral coefficients (LPRHEC) were proposed to detect voice conversion/synthesized signals from genuine speech. The LPRHEC feature captures the magnitude information of LPR signals via short-term spectral analysis.…”

Section: Introductionmentioning

confidence: 99%

“…In fact, the spectral information derived from the original/raw speech signal has the slight imperfection obtained by the distortion of the recording/playback process that may not be effective input of RP feature extraction. Recent studies [24], [25] have indicated that the imperfection produced by recording and playback devices is a crucial clue for improving the performance of feature extraction. Thus, it is expected that the extraction of the RP feature with LPA-based signals may provide better performance than the RP feature using the original/raw speech signal.…”

Section: Introductionmentioning

confidence: 99%

Replay Attack Detection Using Linear Prediction Analysis-Based Relative Phase Features

et al. 2019

View full text Add to dashboard Cite

Recent studies have reported the success of linear prediction analysis (LPA)-related features, which are extracted as a short-term spectral feature for replay attack detection due to the advantage of the imperfection in the LPA-based signal produced by recording and playback devices. However, exploiting LPA-based signals is focused on only magnitude-based features and ignores phase-based features. In this paper, we propose two novel LPA-based relative phase features, namely, linear prediction residual-based relative phase (LPR-RP) and linear prediction analysis estimated speech-based relative phase (LPAES-RP).The key idea of both LPR-RP and LPAES-RP is to extract the phase information based on LPA-based signals. In the LPR-RP feature, we modify the relative phase (RP) feature extraction using a linear prediction residual (LPR) derived from the difference between the original/raw speech and LPA estimated speech signal (LPAES) instead of the original/raw speech signal. LPES-RP feature exploits the LPAES signal to replace the original/raw speech signal. Because the trace of the recording and playback device artifacts is the primary evidence for detecting the replayed signal, the advantages of the imperfection of LPR and LPAES are expected to provide efficient phase information for the replay attack detection task. In addition, using the individual LPR-RP/LPAES-RP feature, our proposed features are combined with two standard features, mel-frequency cepstral coefficients (MFCC), constant Q transform cepstral coefficients (CQCC) and the original RP feature, at score level to further improve the detection decision. The performance of the proposed LPR-RP/LPAES-RP feature and combination are evaluated using the ASVspoof 2017 version 2 database. On the evaluation subset, our proposed LPR-RP and LPAES-RP feature achieves a promising improvement over baseline features (MFCC/CQCC). Moreover, the combined systems of LPR-RP, RP, and CQCC obtains an equal error rate of 9.26%. INDEX TERMSReplay attack detection, phase information, speaker recognition anti-spoofing, linear prediction analysis-based feature.

show abstract

“…Conclusively, the exploration of excitation source information towards replay attacks detection has been ignored by the above mentioned prior works with the exception [9]. LP residual signal as excitation source parameterization has been widely explored already in literature such as in studies [11,12,13,14]. Therefore, excitation source parameterization of LP residual would intuitively be useful for replay attacks detection.…”

Section: Introductionmentioning

confidence: 99%

Linear Prediction Residual based Short-term Cepstral Features for Replay Attacks Detection

Singh

Pati

2018

Interspeech 2018

View full text Add to dashboard Cite

Modern automatic speaker verification (ASV) systems are highly vulnerable to spoof attacks, and developing ASV antispoofing algorithms to protect ASV systems form these attacks is currently a part of active research. Contrarily to current trends on development of stand-alone spoof detection system, this work aims detection of replay attacks directly on the ASV system. The claim made through replay spoofing trials is rejected as impostors directly by ASV system. The objective here is to model the changes in the excitation signal characteristics caused by playback devices for replay detection. Accordingly, two linear prediction (LP) residual based source features are proposed for rejecting replay spoofing trials namely, RMFCC (residual mel-frequency cepstral coefficients) and LPRHEM-FCC (LP residual Hilbert envelope MFCC). A comparative analysis between these two source features has been performed through speaker verification experiments to evaluate their effectiveness for ASV anti-spoofing applications. The comparison between the two has been made in the form of (source feature + MFCC) combination. The experiments are conducted using self-developed IITG-MV replay database. From the experimental results, it has been observed that 'LPRHEMFCC+MFCC' combination outperforms 'RMFCC+MFCC' combination, under replay attacks. Finally, the experiments are repeated on ASVspoof2017 database to validate the efficacy of proposed work.

show abstract

Linear prediction residual features for automatic speaker verification anti-spoofing

Cited by 8 publications

References 34 publications

Evolutionary fusion of classifiers trained on linear prediction based features for replay attack detection

Evolutionary fusion of classifiers trained on linear prediction based features for replay attack detection

Replay Attack Detection Using Linear Prediction Analysis-Based Relative Phase Features

Linear Prediction Residual based Short-term Cepstral Features for Replay Attacks Detection

Contact Info

Product

Resources

About