A comparison of features for synthetic speech detection
Open accessThe DOI matches its Crossref record; 36 of the 36 references it cites were found in a registry or an index.
Abstract
The performance of biometric systems based on automatic speaker recognition technology is severely degraded due to spoofing attacks with synthetic speech generated using different voice conversion (VC) and speech synthesis (SS) techniques. Various countermeasures are proposed to detect this type of at-tack, and in this context, choosing an appropriate feature extrac-tion technique for capturing relevant information from speech is an important issue. This paper presents a concise experi-mental review of different features for synthetic speech detec-tion task. A wide variety of features considered in this study include previously investigated features as well as some other potentially useful features for characterizing real and synthetic speech. The experiments are conducted on recently released ASVspoof 2015 corpus containing speech data from a large number of VC and SS technique. Comparative results using two different classifiers indicate that features representing spectral information in high-frequency region, dynamic information of speech, and detailed information related to subband characteris-tics are considerably more useful in detecting synthetic speech. Index Terms: anti-spoofing, ASVspoof 2015, feature extrac-tion, countermeasures
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“a significant body of research [64,96,97,98,99] has demonstrated that the performance of detection models largely depends on the choice of input features.”
From Audio Deepfake Detection: What Has Been Achieved and What Lies Ahead (2025) · cites this paper as [96]
In the citing paper
“other machine learning models, such as the Support Vector Machine and Logistic Regression, were widely studied in earlier research [96,134,135,136]. With the advancement of feature extraction techniques, these models continue to play a significant role in current studies [137,138,139].”
From Audio Deepfake Detection: What Has Been Achieved and What Lies Ahead (2025) · cites this paper as [96]
Citations
Cites
Multiresolution gray-scale and rotation invariant texture classification with local binary patterns
Ojala, T., Pietikainen, M., & Maenpaa, T. (2002) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2002.1017623 · Cited by 15,398 · cited 1×
“The authors of [8], in turn, proposed to use one-class approach us-ing local binary pattern [9] of linear frequency cepstral coef-ficients (LFCCs) followed by support vector machine (SVM) for voice conversion, speech synthesis and artificial signal detection [8].”
MethodFundamentals of speech recognition
Rabiner, L. R., & Juang, B.-H. F. (1993) · Cited by 7,680
Davis, S., & Mermelstein, P. (1980) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/tassp.1980.1163420 · Cited by 5,465 · cited 1×
“In MFCC, the filters are placed in mel scale, having denser spacing in t he low-frequency region [21].”
MethodPerceptual linear predictive (PLP) analysis of speech
Hermansky, H. (1990) · The Journal of the Acoustical Society of America · doi:10.1121/1.399423 · Cited by 2,536 · cited 1×
“Another all-pole representation of speech called perceptual linear prediction cepstral coefficients (PLPCC) is also computed by first performing a series of perceptual processing prior to LP analysis [24].”
MethodCepstral analysis technique for automatic speaker verification
Furui, S. (1981) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/tassp.1981.1163530 · Cited by 1,173
Continuous probabilistic transform for voice conversion
Stylianou, Y., Cappe, O., & Moulines, E. (1998) · IEEE Transactions on Speech and Audio Processing · doi:10.1109/89.661472 · Cited by 931 · cited 1×
“Synthetic speech signal created using different voice conv ersion (VC) and speech synthesis (SS) techniques can be used to poof biometric systems based on automatic speaker recognition t echnology [1, 2, 3, 4].”
MethodConstruction and evaluation of a robust multifeature speech/music discriminator
Scheirer, E., & Slaney, M. (2002) · 1997 IEEE International Conference on Acoustics, Speech, and Signal Processing · doi:10.1109/icassp.1997.596192 · Cited by 873 · cited 1×
“Spectral flux based feature: Spectral flux measures the frame-by-frame change in the power spectrum [25].”
BackgroundCOVAREP — A collaborative voice analysis repository for speech technologies
Degottex, G., Kane, J., Drugman, T., Raitio, T., & Scherer, S. (2014) · 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2014.6853739 · Cited by 695 · cited 1×
“2, First 20 coefficients are retained after DCT [5] Short-term APGDF (60) LP Order=20 — phase features CosPhase (60) First 20 coefficients are retained after DCT [5] RPS (60) RPS computed with COVAREP tool [14], 20 filters in mel filter ba nk [6, 7]”
MethodSpoofing and countermeasures for speaker verification: A survey
Wu, Z., Evans, N., Kinnunen, T., Yamagishi, J., Alegre, F., & Li, H. (2015) · Speech Communication · doi:10.1016/j.specom.2014.10.005 · Cited by 601
Self-citationThe cepstrum: A guide to processing
Childers, D. G., Skinner, D. P., & Kemerait, R. C. (1977) · Proceedings of the IEEE · doi:10.1109/proc.1977.10747 · Cited by 600 · cited 1×
“Cepstrum: Cepstral coefficients (Cep) are computed from the power spectrum by applying discrete cosine transform (DCT) [17].”
MethodSpectral voice conversion for text-to-speech synthesis
Kain, A., & Macon, M. W. (2002) · Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181) · doi:10.1109/icassp.1998.674423 · Cited by 552
ASVspoof 2015: the first automatic speaker verification spoofing and countermeasures challenge
Wu, Z., Kinnunen, T., Evans, N., Yamagishi, J., Hanilçi, C., Sahidullah, M., & Sizov, A. (2015) · Interspeech 2015 · doi:10.21437/interspeech.2015-462 · Cited by 484 · cited 10×
“The overall best recognition accuracy on development set (EER of 0 .”
BackgroundMethodInfluentialSelf-citationTorres-Carrasquillo, P. A., Singer, E., Kohler, M. A., Greene, R. J., Reynolds, D. A., & Deller Jr., J. R. (2002) · 7th International Conference on Spoken Language Processing (ICSLP 2002) · doi:10.21437/icslp.2002-74 · Cited by 392
On the use of instantaneous and transitional spectral information in speaker recognition
Soong, F. K., & Rosenberg, A. E. (1988) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/29.1598 · Cited by 270
Evaluation of Speaker Verification Security and Detection of HMM-Based Synthetic Speech
De Leon, P. L., Pucher, M., Yamagishi, J., Hernaez, I., & Saratxaga, I. (2012) · IEEE Transactions on Audio, Speech, and Language Processing · doi:10.1109/tasl.2012.2201472 · Cited by 231
Kinnunen, T., Wu, Z.-Z., Lee, K. A., Sedlak, F., Chng, E. S., & Li, H. (2012) · 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2012.6288895 · Cited by 184 · cited 2×
“Synthetic speech signal created using different voice conv ersion (VC) and speech synthesis (SS) techniques can be used to poof biometric systems based on automatic speaker recognition t echnology [1, 2, 3, 4].”
MethodSelf-citationDetecting converted speech and natural speech for anti-spoofing attack in speaker recognition
Wu, Z., Chng, E. S., & Li, H. (2012) · Interspeech 2012 · doi:10.21437/interspeech.2012-465 · Cited by 175 · cited 4×
“Cosine-phase function (CosPhase): Phase spectrum obtained during short-term speech analysis is used for synthe tic speech detection [5].”
MethodInfluentialSpectral subband centroid features for speech recognition
Paliwal, K. K. (2002) · Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181) · doi:10.1109/icassp.1998.675340 · Cited by 147 · cited 1×
“SSFCs are then obtained by performing logarithm and DCT on SSFs. Subband spectral centroid based feature: Spectral subband centroids represent centroid frequencies of subbands, and they have properties similar to formant frequencies [26].”
MethodThe modified group delay function and its application to phoneme recognition
Murthy, H. A., & Gadde, V. R. R. (2003) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2003.1198718 · Cited by 140 · cited 1×
“Modified group delay function (MGDF): Modified group delay function was proposed to represent the phase information of a signal [28].”
MethodSynthetic speech detection using temporal modulation feature
Wu, Z., Xiao, X., Chng, E. S., & Li, H. (2013) · 2013 IEEE International Conference on Acoustics, Speech and Signal Processing · doi:10.1109/icassp.2013.6639067 · Cited by 118 · cited 2×
“Spectral SDC (56) From MFCC withN=7,d=1,P=3,K=7 — features Mod-Spec (60) From20 mel filter log-energies using window of 510 ms with shift of 10 ms [15] with long-term FDLP (60) FDLP package 1 — processing MHEC (60) No.”
MethodKanedera, N., Arai, T., Hynek Hermansky, B., & Misha Pavel, C. (1999) · Speech Communication · doi:10.1016/s0167-6393(99)00002-3 · Cited by 112 · cited 1×
“Modulation spectrum (ModSpec): Modulation spectrum contains long-term temporal characteristics of speech sig nal [31].”
BackgroundAlegre, F., Amehraye, A., & Evans, N. (2013) · 2013 IEEE Sixth International Conference on Biometrics: Theory, Applications and Systems (BTAS) · doi:10.1109/btas.2013.6712706 · Cited by 107
Chakroborty, S., Roy, A., & Saha, G. (2008) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.1330572 · Cited by 91
Toward a Universal Synthetic Speech Spoofing Detection Using Phase Information
Sanchez, J., Saratxaga, I., Hernaez, I., Navas, E., Erro, D., & Raitio, T. (2015) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2015.2398812 · Cited by 88 · cited 8×
“It was used in [6, 7] for detecting synthetic speech.”
MethodInfluentialUTD-CRSS Systems for 2018 NIST Speaker Recognition Evaluation
Zhang, C., Bahmaninezhad, F., Ranjan, S., Dubey, H., Xia, W., & Hansen, J. H. L. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8683097 · Cited by 79
Spoofing countermeasures to protect automatic speaker verification from voice conversion
Alegre, F., Amehraye, A., & Evans, N. (2013) · 2013 IEEE International Conference on Acoustics, Speech and Signal Processing · doi:10.1109/icassp.2013.6638222 · Cited by 77 · cited 1×
“of filter=20 — LPCC (60) LP Order=20 [13] PLPCC (63) No.”
Speaker verification against synthetic speech
Chen, L., Guo, W., & Dai, L.-R. (2010) · International Symposium on Chinese Spoken Language Processing (ISCSLP) · doi:10.1109/iscslp.2010.5684887 · Cited by 66 · cited 1×
“Here, however, we retain all the coefficients since especially the higher-order coeffi-cients could be useful for characterizing synthetic speech [18].”
BackgroundVoice conversion and spoofing attack on speaker verification systems
Wu, Z., & Li, H. (2013) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipa.2013.6694344 · Cited by 63
Classifiers for synthetic speech detection: a comparison
Hanilçi, C., Kinnunen, T., Sahidullah, M., & Sizov, A. (2015) · Interspeech 2015 · doi:10.21437/interspeech.2015-466 · Cited by 57 · cited 1×
“In a different study with classifiers, we have shown that GMM-based technique yields reasonably good accuracy in ASVspoof 2015 corpus [36].”
MethodSelf-citationRobust Feature Extraction Using Modulation Filtering of Autoregressive Models
Ganapathy, S., Mallidi, S. H., & Heřmanský, H. (2014) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2014.2329190 · Cited by 49 · cited 1×
“FDLP features were recently studied in speaker recognition with promising results in both clean and noisy conditions [34].”
BackgroundInvestigation of Spectral Centroid Magnitude and Frequency for Speaker Recognition.
Kua, J. M. K., Thiruvaran, T., Nosratighods, M., Eliathamby Ambikairajah, A., & Epps, J. (2010) · Odyssey · Cited by 47
A new speaker verification spoofing countermeasure based on local binary patterns
Alegre, F., Vipperla, R., Amehraye, A., & Evans, N. (2013) · Interspeech (USB) · doi:10.21437/interspeech.2013-291 · Cited by 45 · cited 1×
“of filter=20 [4, 11, 12] features RFCC/IMFCC (60) No.”
Simple representation of signal phase for harmonic speech models
Saratxaga, I., Hernáez, I., Erro, D., Navas, E., & Sánchez, J. (2009) · Electronics Letters · doi:10.1049/el.2009.3328 · Cited by 44
Mean hilbert envelope coefficients (MHEC) for robust speaker recognition
Sadjadi, S. O., Hasan, T., & Hansen, J. H. L. (2012) · Interspeech (USB) · doi:10.21437/interspeech.2012-464 · Cited by 40 · cited 1×
“Finally, MHEC features are derived by dividing the subband signals into sub-frames and computing the mean [35].”
MethodUsing group delay functions from all-pole models for speaker recognition
Rajan, P., Kinnunen, T., Hanilçi, C., Pohjalainen, J., & Alku, P. (2013) · Interspeech (USB) · doi:10.21437/interspeech.2013-416 · Cited by 39 · cited 1×
“All-pole group delay function (APGDF): Recently, a phase-based feature using all-pole modeling is investigat ed in speaker recognition [29].”
MethodSelf-citationDimension reduction of the modulation spectrogram for speaker verification.
Kinnunen, T., Lee, K. A., & Li, H. (2008) · Cited by 38
Self-citation
Cited by
ASVspoof 2019: Future Horizons in Spoofed and Fake Audio Detection
Todisco, M., Wang, X., Vestman, V., Sahidullah, M., Delgado, H., Nautsch, A., Yamagishi, J., Evans, N., Kinnunen, T., & Lee, K. A. (2019) · arXiv · doi:10.48550/arxiv.1904.05441 · Cited by 688
“ms. They use a common Gaussian mixture model (GMM) back-end classifier with either constant Q cepstral coefficient (CQCC) features [22, 23] (B01) or linear frequency cepstral coefficient (LFCC) features [24] (B02). 4. Challenge results Table 1 shows results4 in terms of the t-DCF and EER for primary systems, pooled over…”
MethodSelf-citationASVspoof 2019: Future Horizons in Spoofed and Fake Audio Detection
Todisco, M., Wang, X., Vestman, V., Sahidullah, M., Delgado, H., Nautsch, A., Yamagishi, J., Evans, N., Kinnunen, T. H., & Lee, K. A. (2019) · Interspeech 2019 · doi:10.21437/interspeech.2019-2249 · Cited by 687
Self-citationASVspoof 2019: A large-scale public database of synthesized, converted and replayed speech
Wang, X., Yamagishi, J., Todisco, M., Delgado, H., Nautsch, A., Evans, N., Sahidullah, M., Vestman, V., Kinnunen, T., Lee, K. A., et al. (2020) · Computer Speech & Language · doi:10.1016/j.csl.2020.101114 · Cited by 505
“Both use a common Gaussian mixture model (GMM) back-end classifier with either constant-Q cepstral coefficient (CQCC) features [64, 65] (B01) or linear frequency cepstral coefficient (LFCC) features [66] (B02).”
MethodSelf-citationTrends in audio signal feature extraction methods
Sharma, G., Umapathy, K., & Krishnan, S. (2020) · Applied Acoustics · doi:10.1016/j.apacoust.2019.107020 · Cited by 479
Constant Q Cepstral Coefficients: A Spoofing Countermeasure for Automatic Speaker Verification
Todisco, M., Delgado, H., & Evans, N. (2017) · Computer Speech & Language · doi:10.1016/j.csl.2017.01.001 · Cited by 424
Metadata
- Authors
- Sahidullah, M., Kinnunen, T., & Hanilçi, C.
- Affiliations
- University of Eastern Finland (FI)
- Venue
- Interspeech 2015
- Publisher
- ISCA
- Published
- 2015-09-06
- Pages
- 2087-2091
- ISSN
- 2308-457X (ISSN-L)
- OpenAlex
- W2295634712
- Cited by
- 382
- References
- 36
- Citations by year
- 2026: 23 · 2025: 41 · 2024: 42 · 2023: 50 · 2022: 38 · 2021: 51 · 2020: 35 · 2019: 41 · 2018: 15 · 2017: 29 · 2016: 15 · 2015: 2
- FWCI
- 11.46
- Type
- conference paper
- Language
- en
- Open access
- green
- erepo.uef.fi
- Licence
- other-oa
- Locations
- Interspeech (USB) (publishedVersion)
- UEF eRepo (University of Eastern Finland) (submittedVersion, other-oa)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Physical Sciences › Computer Science › Signal Processing › Biometric Identification and Security
- Keywords
- synthetic speech detection, spoofing attacks, anti-spoofing, speaker verification, voice conversion, speech synthesis, feature extraction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.