Conceptio › Archive › arXiv CS
arXiv CSopen access

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System

2026 · arxiv_cs
arXiv CS · Papers · License: Open Access · 2026
Open Source ↗Direct PDF ↓
cryptographycybersecurityprivacysecurity
cryptography, security, privacy, cybersecurity

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System Phuc-Khang Vo-Hoang⋆† Hoang C. Ta⋆† Nhien-An Le-Khac‡ † Dinh-Thuc Nguyen Hong-Hanh Nguyen-Le‡ University of Science, Ho Chi Minh City, Vietnam University College Dublin, School of Computer Science, Dublin, Ireland

arXiv:2605.02990v1 [cs.CR] 4 May 2026

†

‡

Abstract. In this work, we present a Challenge-Response Voice Cancelable authentication system, called ChaRVoC, which provides protection against replay attacks, revocability issues, and template compromise. Our approach integrates three security factors: (1) inherent voice biometric characteristics, (2) user-memorized secret keys enabling template revocability, and (3) dynamic system-generated challenges providing liveness detection. Specifically, we introduce a novel HashGray-XOR scheme which combines a cryptographic hash function with an unrecoverable graycode-based transformation to create secured templates that are mathematically proven to be non-invertible. We compare our methods with existing cancelable biometric methods (WTA, IoM, RoE) on VoxCeleb1, TIMIT, and VOiCES datasets to show the recognition performance of our proposed system. We also show that our system achieves both cancelability and unlinkability properties. Keywords: Challenge-response authentication · voice biometric · cancelable biometric

1

Introduction

Voice-based authentication systems have gained remarkable popularity in recent years, driven by the proliferation of voice-controlled digital assistants, smart home devices, and mobile applications [24,10]. However, current voice authentication systems [4,19,18,21] suffer from three fundamental vulnerabilities that limit their deployment in practice. First, voice biometrics are permanently linked to users and cannot be revoked if compromised [1]. Second, stored voice templates in databases present targets for attackers, as a single breach can compromise users across multiple systems [4,19]. Third, these systems are vulnerable to replay attacks, where recorded voice samples can be used to impersonate legitimate users [22]. While cancelable biometric techniques have been proposed to protect templates stored in databases, existing approaches either require complex cryptographic operations that are computationally expensive [2,18,13] or they fail to ⋆

Two authors have the equal contribution

2

Phuc-Khang Vo-Hoang et al.

address the replay attack vulnerability [22]. Text-dependent speaker verification systems can prevent replay attacks through dynamic phrases [11,5], but they cannot prevent voice biometrics stored in databases from server-specific attacks. Current solutions address these challenges in isolation [16,26,19,18], resulting in systems that remain vulnerable to at least one attack vector. In this work, we present a Challenge-Response Voice Cancelable authentication system, named ChaRVoC, which simultaneously addresses replay attacks, revocability issues, and template compromise. Our approach combines three security factors: (1) the user’s voice biometric characteristics, (2) a user-specific secret key (i.e., PIN) that each person memorizes, and (3) a system-generated random challenge that provides replay attack resistance. Critically, we introduce our novel HashGray-XOR scheme that combines the secret key with the voice features to create protected templates that are both cancelable and non-invertible. Particularly, this scheme consists of two functions: (i) a cryptographic hash function that maps the random numeric sequence to a fixed-length binary string and (ii) an unrecoverable graycode-based function that transforms a real-valued voice feature vector into a binary representation. The binary feature and the hashed key are combined using the XOR operation to create the final protected template. We also mathematically prove the non-invertibility of our HashGray-XOR and demonstrate that templates generated from our system are computationally unlinkable. Read this random number, please:

198765

Secret key

Speaker extractor

Graycode-based function

Cryptographic hash function

XoR

S2T

Yes

Matching

Yes

Check

No

No

Reject

Reject

Accept

Fig. 1: An overview of our Challenge-Response Voice Cancelable Voice Authentication system, namely ChaRVoC.

2

Related Work

Existing voice-based authentication systems address replay attacks and template compromise in isolation. Text-dependent speaker verification systems [11,5] mitigate replay attacks by requiring users to speak dynamically generated phrases. Zhange et al. [27] detect replay attacks through analysis of articulatory movements during speech production, distinguishing genuine human speech from loudspeaker reproductions based on physical mouth and tongue dynamics. To

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System

3

protect voice biometrics stored in databases, cryptographic techniques are applied, including Paillier cryptosystem [18], fuzzy commitment scheme [2], and fuzzy vault scheme [13]. Alternatively, non-invertible transformation approaches provide template security through mathematical one-way functions. El-Moneim et al. [7] achieve template revocability by transforming speech into spectrograms and extracting user-key-dependent patches. Nguyen et al. [20] proposed a Ranking-of-Element hashing to encode elements based on the count of smallervalued elements. Unlike these isolated solutions, we propose a unified framework addressing replay attacks, template compromise, and revocability simultaneously. While Ceaparu et al. [3] present a similar multi-factor approach, they require preregistered personal smartphones, whereas our system automatically generates challenge sequences without additional device requirements, improving deployment flexibility.

3

Proposed Method

3.1

System Overview

As illustrated in Figure 1, the ChaRVoC system operates through several key components to provide multi-layered security. The Random Challenge Generator produces system-generated numeric sequences (e.g., "198765") that serve as dynamic authentication challenges. These challenges change with each authentication attempt, ensuring that no two authentication sessions use the same input. The speech-to-text (S2T) module verifies that the user correctly speaks the displayed random number, providing real-time liveness detection and preventing replay attacks. Even if an attacker obtains a recording of the user’s voice, they cannot authenticate without speaking the current challenge. Simultaneously, a speaker extractor R transforms raw voice samples into discriminative real-valued feature vectors, capturing the unique voice characteristics of the speaker. The core security mechanism of our system is the proposed HashGrayXoR scheme, which protects biometric templates through two integrated functions: the graycode-based function T and the cryptographic hash function H. During enrollment, the user provides a voice sample and a secret key. A speaker extractor R extracts features from the voice sample, which are transformed into a binary template via the graycode-based function. Simultaneously, the secret key is hashed using H to produce a binary string. These are combined through XOR to create the protected template. During authentication, the system displays a random challenge that the user reads aloud while providing their secret key. The spoken response is processed through two parallel paths: the S2T module verifies that the correct challenge was spoken, and the speaker recognition model extracts voice features. These features are transformed into a binary template via the graycode-based function T . Meanwhile, the provided secret key is hashed using H, and this hash is XORed with the stored protected template to recover what should be the original enrollment template. This recovered template is then matched against the current voice template. Authentication suc-

4

Phuc-Khang Vo-Hoang et al.

ceeds only when the S2T verification passes, and the recovered template matches the current voice template. 3.2

HashGray-XOR Scheme

The HashGray-XOR scheme is formally defined as a transformation f : {0, 1}∗ × I → {0, 1}n that takes as input a secret key k ∈ {0, 1}∗ and a voice sample v ∈ I with I, where I is the set of all possible voice samples. The output is an n-bit binary hash value t ∈ {0, 1}n , defined by: t = f (k, v) = H(k) ⊕ T (v), ∀k ∈ {0, 1}k , v ∈ I,

(1)

where H : {0, 1}∗ → {0, 1}n is the cryptographic hash function, T : I → {0, 1}n is the unrecoverable graycode-based function, and ⊕ denotes the bitwise XOR operation. Unrecoverable Graycode-Based Function The unrecoverable function T transforms real-valued voice features into binary representations through a composition function: T (v) = G ◦ r ◦ R(v) = G(r(R(v))), where G : N → {0, 1}l+1 is the graycode encoding function, r : R → Z is the roundoff function defined as r(x) = round(x × 10p ) with precision parameter p. The sign of each voice feature vi = r(fi ) after being applied to the roundoff function r is encoded as a binary bit b0 = 0 if vi < 0 and b0 = 1 if vi ≥ 0. The integer value |vi | is converted to an (l + 1)-bit Graycode representation. Theorem 1. (Unrecoverability Property) The function T : I → 0, 1n constructed as above is unrecoverable, meaning that with t = T (v), it is computationally infeasible to find back the original v ∈ I. To prove this theorem, we establish two lemmas: Lemma 1. (Non-invertibility of Roundoff Function) The function r : R → Z defined by ( ⌊x⌋, if |x − ⌊x⌋| < 0.5 r(x) = ⌈x⌉, otherwise is not invertible. Indeed, if r is invertible, then r is injective, meaning that ∀x1 , x2 , if x1 ̸= x2 then r(x1 ) ̸= r(x2 ). This is contradicted by the fact that r(3.6) = r(3.9) = 4 but 3.6 ̸= 3.9. Lemma 2. Given f = h ◦ g, the function f is invertible if and only if h and g are invertible functions.

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System

5

Proof. (Lemma 2) To prove that f is invertible if and only if g and h are invertible functions, we need to prove two parts. Part 1: If g and h are invertible functions. Then there exist inverse functions g −1 and h−1 . The composite function f = h ◦ g means f (x) = h(h(x)). To prove that f is invertible, we need to find the inverse function f −1 . We have f −1 = h−1 ◦ g −1 (by the property of composite functions). Indeed, −1 (f ◦ f )(x) = (h−1 ◦ g −1 )(x) = x. Similarly, (f ◦ f −1 )(x) = x. Therefore, f is invertible. Part 2. If f is invertible, then g and h are invertible. Assume f is invertible, then there exists an invertible function f −1 . The composite function f = g ◦ h means f (x) = g(h(x)). We need to prove that g and h are invertible. For g, suppose g is not invertible. There exists x1 ̸= x2 such that g(x1 ) = g(x2 ). Therefore, f (x1 ) = g(h(x1 )) = g(h(x2 )) = f (x2 ). This contradicts the fact that f is invertible (because f (x1 ) ̸= f (x2 ) when x1 ̸= x2 ). Therefore, g must be invertible. Similarly, h is proved as the same. Conclusion. f is invertible if and only if g and h are invertible functions. Proof. (Theorem 1) For any v ∈ I, let b = (b0 , b1 , ..., bl ) = T (v) = G(r(R(v))). Suppose that given b = (b0 , b1 , ..., bl ) = T (v) = G(r(R(v))), one can recover exactly v : T (v) = b. This means given b = (b0 , b1 , ..., bl ), there is only one v : T (v) = b, or T = G ◦ H is invertible, where H = r ◦ R. By Lemma 2, H = r ◦R is invertible if and only if r and R are invertible functions. This contradicts the fact that r is not invertible (by Lemma 1). Therefore, T is unrecoverable. Cryptographic Hash Function The cryptographic hash function H is defined as: H : {0, 1}∗ → {0, 1}n . This function maps an arbitrary binary string (a secret key) to a fixed-size binary string of n bits, serving as the biometric template protection mechanism. Definition 1. (Cancelable Biometric Function) A biometric function is cancelable if it can be designed to allow an individual to enroll and revoke many different samples. And the value of the biometric function for a generated sample is called a biometric template. Definition 2. A cryptographic hash function H : {0, 1}∗ → {0, 1}n has the following desirable properties: 1. Pre-image resistance: Given a hash value y, it is computationally difficult to find any x such that y = H(x). 2. Second pre-image resistance: Given a pre-image x1 , it is computationally difficult to find another pre-image x2 such that H(x1 ) = H(x2 ). 3. Collision resistance: It is computationally difficult to find two different values x1 and x2 such that H(x1 ) = H(x2 ).

6

Phuc-Khang Vo-Hoang et al.

Theorem 2. (Cancelability Property) Given a cryptographic hash function H : {0, 1}∗ → {0, 1}n and an unrecoverable function of voice biometric T : I → {0, 1}n , the voice biometric function defined as follows is cancelable: f : {0, 1}∗ × I → {0, 1}n defined by f (k, v) = H(k) ⊕ T (v), ∀k ∈ {0, 1}∗ , v ∈ I. Proof. First, we note that if t1 = H(k1 ) ⊕ T (v1 ) and t2 = H(k2 ) ⊕ T (v2 ) are two voice biometric templates of the same person at two different sampling times, v1 and v2 , we have:  0, if k1 = k2 , v1 = v2    T (v ) ⊕ T (v ), if k1 = k2 , v1 ̸= v2 1 2 t1 ⊕ t2 =  H(k ) ⊕ H(k ), if k1 ̸= k2 , v1 = v2 1 2    (H(k1 ) ⊕ H(k2 )) ⊕ (T (v1 ) ⊕ T (v2 )), otherwise Theorem 2 shows that even if the voice biometric samples have the same binary representation but different secret keys, the voice biometric function values will be different with high probability if the cryptographic hash function has high collision resistance. Theorems 1 and 2 ensure that our ChaRVoC system provides both noninvertible and unrecoverable template protection. 3.3

Template Matching

The matching process compares binary templates generated during enrollment and authentication. For two binary templates T1 = b10 b11 ...b1n and T2 = b20 b21 ...b2n , the similarity measure is: Pn (1 − |b1i − b2i |) i=0 Pn (2) S(T1 , T2 ) = 2n − i=0 (1 − |b1i − b2i |) This computes the ratio of matching bits, producing a similarity score between 0 and 1, where 1 indicates identical templates and 0 indicates completely different templates. Fig. 2: Distributions of mated samples and non-mated samples.

(a) VoxCeleb1

(b) TIMIT

(c) VOiCES

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System

7

Table 1: Performance Comparison of Different Speaker Recognition Models Across Three Datasets. Lower EER and Higher AUC/TMR Values Indicate Better Performance. Dataset

Metric

Model EER

AUC TMR@FMR=0.1%

VoxCeleb1

RawNet3 5.0901 0.9836 RawNet2 2.8791 0.9953 ECAPA-TDNN 11.0869 0.9564

0.7588 0.7509 0.4045

TIMIT

RawNet3 2.4901 0.9854 RawNet2 1.7008 0.9986 ECAPA-TDNN 7.0233 0.9819

0.9272 0.8786 0.5774

VOiCES

RawNet3 26.1745 0.7796 RawNet2 11.3680 0.9537 ECAPA-TDNN 11.6902 0.9484

0.0250 0.4774 0.6016

4

Experiments

4.1

Experimental Setup

Datasets. We use three datasets: VoxCeleb1 [17], TIMIT [8], and VOiCES [23] to evaluate our proposed ChaRVoC system. VOiCES is utilized to evaluate the accuracy performance of our system in noisy conditions (e.g., background noise and reverberation). Speaker extractors. To demonstrate the compatibility of our HashGrayXOR scheme, we utilize three popular speaker recognition models, including ECAPA-TDNN [6], RawNet2 [14], and RawNet3 [15]. For all models, we employ publicly available pretrained weights. Baselines. We compare our proposed method with three cancellable biometrics approaches, Winner-Take-All (WTA) [25], Index-of-Max (IoM) [12] hashing, and Ranking-of-Element (RoE) hashing [19]. Metrics. We use three evaluation metrics: equal error rate (EER), the area under the ROC curve (AUC), and True Match Rate at a False Match Rate of 0.1% (TMR@FMR=0.1%). 4.2

Experimental Results

This section presents the results of our ChaRVoC system regarding recognition performance and computation efficiency. Furthermore, we also analyze the cancelability and unlinkability of our system. Recognition Performance. Table 1 shows the performance of our ChaRVoC system with different speaker extractors R. RawNet2 consistently achieves the best overall performance across datasets, with the lowest EER of 2.88% on VoxCeleb1, 1.70% on TIMIT, and 11.368% on the noisy VOiCES dataset.

8

Phuc-Khang Vo-Hoang et al.

Based on these results, we select RawNet2 as our primary speaker extractor for subsequent experiments. Comparison with Baselines. Table 2 demonstrates that our method significantly outperforms existing cancelable biometric approaches across all datasets. Unlike WTA and IoM, which may lose essential discriminative information during the hashing process, our approach preserves the biometric characteristics. Compared to the state-of-the-art voice-based cancelable biometrics system RoE, our method also outperforms with the EER of 2.879%, 1.700%, and 11/368% on VoxCeleb1, TIMIT, and VOiCES datasets, respectively.

Table 2: Performance comparison of our ChaRVoC method against baselines (WTA, IoM, RoE). Lower EER and higher AUC/TMR values indicate better performance. Metric

Dataset Method

EER AUC TMR@FMR=0.1% VoxCeleb1

WTA IoM RoE Our

9.341 0.798 8.260 0.856 3.213 0.959 2.879 0.995

0.576 0.631 0.713 0.751

TIMIT

WTA IoM RoE Our

8.624 0.806 7.256 0.891 2.294 0.976 1.700 0.998

0.8205 0.8384 0.8491 0.878

VOiCES

WTA IoM RoE Our

18.041 0.578 15.297 0.687 11.879 0.894 11.368 0.953

0.301 0.398 0.461 0.477

Time Complexity Evaluation. Table 3 presents a comparison of the time complexity for template generation between our ChaRVoC system and the baseline methods (WTA, IoM, and RoE) on the VoxCeleb1 and TIMIT datasets. Our method demonstrates superior computational efficiency, with template generation times of just 0.0054 seconds on VoxCeleb1 and 0.0052 seconds on TIMIT, significantly faster than all the baselines. Security Analysis. To evaluate unlinkability, we used the framework proposed by Gomez-Barrero et al. [9]. Figure 2 illustrates the similarity score distributions between mated (same voice, different keys) and non-mated (different voices) samples. The substantial overlap between these distributions across all datasets confirms that templates generated from the same voice but different keys are computationally indistinguishable from those of different users, ensuring strong privacy protection against cross-matching attacks.

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System

9

Table 3: Comparison of Time Complexity (in seconds) for Template Generation between ChaRVoC and Baselines.

5

Method

Dataset

Time complexity (s)

WTA [24]

VoxCeleb1 TIMIT

0.0380 0.0380

IoM [25]

VoxCeleb1 TIMIT

0.1050 0.1070

RoE [4]

VoxCeleb1 TIMIT

0.1060 0.1010

ChaRVoC (Ours)

VoxCeleb1 TIMIT

0.0054 0.0052

Conclusion

In this work, we propose ChaRVoC, a Challenge-Response Voice Cancelable authentication system. By integrating voice biometrics, secret keys, and dynamic challenge-response mechanisms, our system provides template revocability, replay attack prevention, and non-invertible protection through the novel HashGray-XOR scheme. Experimental results across VoxCeleb1, TIMIT, and VOiCES datasets demonstrate superior performance of our system over existing cancelable biometric systems, achieving EERs as low as 1.70%. Furthermore, our security analysis confirms strong template unlinkability, while practical processing times (0.76s enrollment, 3.18s authentication) validate real-world deployability.

References 1. Bhargav-Spantzel, A., Squicciarini, A., Bertino, E.: Privacy preserving multi-factor authentication with biometrics. In: Proceedings of the second ACM workshop on Digital identity management. pp. 63–72 (2006) 2. Billeb, S., Rathgeb, C., Reininger, H., Kasper, K., Busch, C.: Biometric template protection for speaker recognition based on universal background models. IET Biometrics 4(2), 116–126 (2015) 3. Ceaparu, M., Toma, S.A., Segarceanu, S., Suciu, G., Gavat, I.: Multifactor voicebased authentication system. J. Eng. Sci. Technol. Rev pp. 131–136 (2020) 4. Chee, K.Y., Jin, Z., Cai, D., Li, M., Yap, W.S., Lai, Y.L., Goi, B.M.: Cancellable speech template via random binary orthogonal matrices projection hashing. Pattern Recognition 76, 273–287 (2018) 5. rahman Chowdhury, F.R., Wang, Q., Moreno, I.L., Wan, L.: Attention-based models for text-dependent speaker verification. In: 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). pp. 5359–5363. IEEE (2018) 6. Desplanques, B., Thienpondt, J., Demuynck, K.: ECAPA-TDNN: emphasized channel attention, propagation and aggregation in TDNN based speaker verification. In: Meng, H., Xu, B., Zheng, T.F. (eds.) Interspeech 2020. pp. 3830–3834. ISCA (2020)

10

Phuc-Khang Vo-Hoang et al.

7. El-Moneim, S.A., Nassar, M.A., Dessouky, M.I., Ismail, N.A., El-Fishawy, A.S., Abd El-Samie, F.E.: Cancellable template generation for speaker recognition based on spectrogram patch selection and deep convolutional neural networks. International Journal of Speech Technology 25(3), 759–770 (2022). https://doi.org/10. 1007/s10772-020-09791-y 8. Garofolo, J.S., et al.: Timit acoustic-phonetic continuous speech corpus ldc93s1. Web Download (1993), https://catalog.ldc.upenn.edu/LDC93S1 9. Gomez-Barrero, M., Galbally, J., Rathgeb, C., Busch, C.: General framework to evaluate unlinkability in biometric template protection systems. IEEE Transactions on Information Forensics and Security 13(6), 1406–1420 (2017) 10. Guamán, S., Calvopiña, A., Orta, P., Tapia, F., Yoo, S.G.: Device control system for a smart home using voice commands: A practical case. In: Proceedings of the 2018 10th International Conference on Information Management and Engineering. pp. 86–89 (2018) 11. Heigold, G., Moreno, I., Bengio, S., Shazeer, N.: End-to-end text-dependent speaker verification. In: 2016 IEEE international conference on acoustics, speech and signal processing (ICASSP). pp. 5115–5119. IEEE (2016) 12. Jin, Z., Hwang, J.Y., Lai, Y.L., Kim, S., Teoh, A.B.J.: Ranking-based locality sensitive hashing-enabled cancelable biometrics: Index-of-max hashing. IEEE Transactions on Information Forensics and Security 13(2), 393–407 (2017) 13. Johnson, R., Boult, T.E.: With vaulted voice verification my voice is my key. In: 2013 IEEE International Conference on Technologies for Homeland Security (HST). pp. 453–459. IEEE (2013) 14. Jung, J.w., Kim, S.b., Shim, H.j., Kim, J.h., Yu, H.J.: Improved rawnet with feature map scaling for text-independent speaker verification using raw waveforms. Proc. Interspeech pp. 3583–3587 (2020) 15. Jung, J.w., Kim, Y.J., Heo, H.S., Lee, B.J., Kwon, Y., Chung, J.S.: Pushing the limits of raw waveform speaker recognition. Proc. Interspeech (2022) 16. Mittal, G., Jakobsson, A., Marshall, K.O., Hegde, C., Memon, N.: Pitch: Aiassisted tagging of deepfake audio calls using challenge-response. arXiv preprint arXiv:2402.18085 (2024) 17. Nagrani, A., Chung, J.S., Zisserman, A.: VoxCeleb: A large-scale speaker identification dataset. arXiv:1706.08612 (2017), available at http://www.robots.ox.ac. uk/~vgg/data/voxceleb/ 18. Nautsch, A., Isadskiy, S., Kolberg, J., Gomez-Barrero, M., Busch, C.: Homomorphic encryption for speaker recognition: Protection of biometric templates and vendor model parameters. arXiv preprint arXiv:1803.03559 (2018) 19. Nguyen-Le, H.H., Tran, L., Nguyen, D.S.A., Le-Khac, N.A., Nguyen, T.: Privacypreserving speaker verification system using ranking-of-element hashing. Pattern Recognition 159, 111107 (2025) 20. Nguyen-Le, H.H., Tran, L., Nguyen, D.S.A., Le-Khac, N.A., Nguyen, T.: Privacypreserving speaker verification system using ranking-of-element hashing. Pattern Recognition 159, 111107 (2025). https://doi.org/10.1016/j.patcog.2024. 111107 21. Paulini, M., Rathgeb, C., Nautsch, A., Reichau, H., Reininger, H., Busch, C.: Multibit allocation: Preparing voice biometrics for template protection. In: Odyssey. pp. 291–296 (2016) 22. Pradhan, S., Sun, W., Baig, G., Qiu, L.: Combating replay attacks against voice assistants. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 3(3), 1–26 (2019)

ChaRVoC: A Challenge-Response Voice Cancelable Authentication System

11

23. Richey, C., Barrios, M.A., Armstrong, Z., Bartels, C., Franco, H., Graciarena, M., Lawson, A., Nandwana, M.K., Stauffer, A., van Hout, J., Gamble, P., Hetherly, J., Stephenson, C., Ni, K.: Voices obscured in complex environmental settings (voices) corpus (2018) 24. Yadav, S.P., Gupta, A., Nascimento, C.D.S., de Albuquerque, V.H.C., Naruka, M.S., Chauhan, S.S.: Voice-based virtual-controlled intelligent personal assistants. In: 2023 International Conference on Computational Intelligence, Communication Technology and Networking (CICTN). pp. 563–568. IEEE (2023) 25. Yagnik, J., Strelow, D., Ross, D.A., Lin, R.s.: The power of comparative reasoning. In: 2011 International Conference on Computer Vision. pp. 2431–2438. IEEE (2011) 26. Yasur, L., Frankovits, G., Grabovski, F.M., Mirsky, Y.: Deepfake captcha: A method for preventing fake calls. In: Proceedings of the 2023 ACM Asia Conference on Computer and Communications Security. pp. 608–622 (2023) 27. Zhang, L., Tan, S., Yang, J.: Hearing your voice is not enough: An articulatory gesture based liveness detection for voice authentication. In: Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security. pp. 57–71 (2017)

Record · ID 157320 · SHA-256 e09771a4262e4161
Retrieved via Conceptio — every document is proof-bundled with source, license, and retrieval metadata.