Securing the voice channel: Advanced voice biometric authentication for contact centers

 Corresponding author: Gokulkumar Selvanathan. Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution License 4.0. Securing the voice channel: Advanced voice biometric authentication for contact centers Gokulkumar Selvanathan * Alagappa University, India. World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 Publication history: Received on 07 April 2025; revised on 19 May 2025; accepted on 21 May 2025 Article DOI: https://doi.org/10.30574/wjarr.2025.26.2.1950 Abstract Voice biometrics is emerging as a transformative technology for contact centers seeking enhanced security without compromising customer experience. This comprehensive article examines how voice authentication leverages the unique vocal characteristics of individuals to create secure identity verification processes. By exploring the fundamental components of voice biometric systems—from enrollment and matching to liveness detection and anti-spoofing measures—it provides insights into the technological underpinnings of this authentication method. The article details integration approaches through APIs and decisioning frameworks that enable seamless implementation within existing customer journeys. It addresses key challenges including environmental factors, voice variations, and aging effects on voice patterns. The discussion concludes with an examination of privacy considerations and regulatory compliance requirements, demonstrating how voice biometrics, when implemented thoughtfully, offers a balanced solution that strengthens security while eliminating the friction associated with traditional authentication methods. Keywords: Voice Biometrics; Contact Center Security; Frictionless Authentication; Voiceprint Verification; AntiSpoofing Technology 1. Introduction to Voice Authentication in Contact Centres 1.1. Current Authentication Challenges The modern contact center faces a security paradigm that has evolved dramatically in recent years. Call volumes continue to rise, with the average Fortune 500 contact center now handling approximately 50,000 calls daily according to Bhattacharyya and Sanyal's comprehensive performance evaluation of behavioral biometric systems [1]. This escalating interaction volume correlates with increased fraud attempts, as traditional authentication methods demonstrate significant vulnerabilities. Knowledge-based authentication (KBA) systems show false rejection rates between 10-15% for legitimate customers while maintaining susceptibility to social engineering tactics [1]. Passwordbased systems fare no better, with compromise rates estimated at 27.9% across multiple security breach analyses [2]. 1.2. Voice Biometrics as a Strategic Solution Voice biometric technology represents a sophisticated approach to these authentication challenges by leveraging the approximately 100 distinct physiological and behavioral characteristics present in human speech. The technology's effectiveness is evident in its statistical performance metrics, with enterprise implementations demonstrating equal error rates (EER) between 1.2% and 4.7% depending on environmental factors and implementation maturity [2]. Kumar and Chen's extensive analysis reveals that text-independent voice biometric systems in particular have shown remarkable resilience against replay attacks, with advanced systems detecting such attempts with 97.3% accuracy [2]. The technology's dual capacity to enhance security while reducing friction has driven market adoption, with voice World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3212 biometrics implementation increasing 34.8% year-over-year across financial services, healthcare, and telecommunications sectors [1]. 1.3. Operational and Financial Implications The operational impact of voice biometric implementation extends beyond security enhancements. Organizations transitioning from traditional methods report average handling time reductions of 42 seconds per authentication event, translating to approximately $0.78 per call in operational savings [1]. When scaled across enterprise contact center volumes, Kumar and Chen calculate this represents potential annual savings between $3.2 million and $5.7 million for organizations handling over 20 million calls annually [2]. Furthermore, customer experience metrics show consistent improvement, with post-implementation Net Promoter Score (NPS) increases averaging 18 points across multiple industry case studies analyzed by Bhattacharyya and Sanyal [1]. These improvements reflect the elimination of friction points that previously required customers to recall complex passwords or personal information during high-stress service interactions. 2. Understanding Voice as a Biometric Identifier 2.1. Acoustic Fundamentals of Voiceprints Human voice contains distinct acoustic patterns that form the foundation for biometric identification. According to comprehensive research by Martinez et al., these patterns derive from both physiological structures and behavioral characteristics, creating a multilayered biometric identifier with approximately 547 measurable features [3]. The physiological components—including vocal tract length (averaging 17.5cm in adult males and 14.1cm in females) and laryngeal cavity dimensions—contribute to fundamental frequency variations that range from 85-155Hz for adult males and 165-255Hz for adult females [4]. These inherent physical differences create baseline acoustic signatures that remain relatively stable over time. Complementing these physiological factors are behavioral elements that manifest through linguistic patterns, pronunciation tendencies, and speech cadence, which contribute an additional layer of uniqueness to individual voice patterns and enhance the discriminative power of voice as a biometric identifier [3]. 2.2. Signal Processing and Feature Extraction Modern voice biometric systems leverage sophisticated signal processing techniques to isolate and analyze identification markers. The predominant approach employs mel-frequency cepstral coefficients (MFCCs), which divide speech signals into frames of approximately 20-30 ms with 10 ms overlaps, creating 100 frames per second of speech [4]. From each frame, systems typically extract 12-19 primary coefficients plus a log-energy parameter, resulting in feature vectors with dimensionality between 24-60 when including first and second derivatives [3]. These measurements undergo Gaussian Mixture Model (GMM) processing, creating voiceprint templates averaging 7.2 KB in size—significantly smaller than other biometric templates such as facial recognition (15-20 KB) or iris scans (10-12 KB) [4]. This efficiency in template size enables rapid processing, with modern systems capable of completing verification in 1.2 seconds while maintaining equal error rates (EERs) of 0.72% under ideal conditions and 2.83% in typical operational environments [3]. 2.3. Comparative Accuracy and Performance Metrics When benchmarked against other biometric modalities, voice verification demonstrates a distinct performance profile. Recent studies analyzing 23 biometric systems across 12 authentication contexts found that fingerprint recognition maintains marginally higher accuracy with EERs of 0.5% versus voice biometrics' 2.1% in controlled environments [3]. However, voice authentication offers significant advantages in user acceptance metrics, with implementation satisfaction ratings of 87.6% compared to 72.3% for fingerprint and 63.8% for facial recognition [4]. The non-intrusive nature of voice collection substantially improves user experience, particularly evident in the 42% reduction in authentication abandonment rates observed across multiple enterprise implementations [3]. Furthermore, voice biometrics demonstrates remarkable anti-spoofing capabilities, with current systems detecting synthetic speech attacks with 97.2% accuracy and replay attacks with 94.5% effectiveness when incorporating liveness detection measures [4]. World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3213 Table 1 Comparison of Voice Biometric Performance Metrics Across Authentication Methods [3, 4] Authentication Method Authentication Time (seconds) False Acceptance Rate (FAR) User Satisfaction Rating (%) Text-Dependent Voice 2.3 0.01% 87.6% Text-Independent Voice 3.7 0.04% 92.1% Knowledge-Based Questions 12.8 4.2% 63.8% Password Authentication 8.4 10.1% 71.3% 3. Core Components of Voice Biometric Systems 3.1. Enrollment and Template Generation The foundation of any voice biometric system lies in its enrollment process, which establishes the baseline template against which future authentication attempts are compared. According to comprehensive research by Shrivastava and colleagues, effective enrollment requires capturing between 15-30 seconds of speech to generate statistically significant templates. Their analysis of 324 enrollment sessions across diverse demographic groups revealed that multi-session enrollment protocols—collecting 3-4 samples of 5-7 seconds each over separate interactions—improved subsequent authentication accuracy by 18.6% compared to single-session approaches. The resulting voiceprint templates typically occupy 3-5 KB of storage and contain mathematical representations of approximately 38-42 distinct voice parameters extracted through Mel-Frequency Cepstral Coefficient (MFCC) analysis performed across 20 ms frames with 10 ms overlapping windows [5]. These templates must balance comprehensiveness against computational efficiency, with research demonstrating that templates incorporating additional prosodic features beyond spectral characteristics reduced equal error rates (EER) from 4.2% to 2.7% at the cost of increasing template size by approximately 22% [6]. 3.2. Matching Algorithms and Performance Optimization The core verification process employs sophisticated algorithms that compare incoming voice samples against stored templates to determine identity. Contemporary systems predominantly utilize either Gaussian Mixture Models (GMM) with Universal Background Models (UBM) or i-vector approaches with Probabilistic Linear Discriminant Analysis (PLDA) for comparison operations. These models operate by converting speech into feature vectors that typically contain 12-19 coefficients per frame, with normalization techniques applied to mitigate channel effects. Experiments with 578 verification attempts under varying conditions demonstrated that i-vector/PLDA systems achieved EERs of 1.8% under ideal conditions and 3.9% in environments with signal-to-noise ratios of 15 dB [6]. Threshold configuration represents a critical operational consideration, with Shrivastava's analysis revealing that financial service implementations typically calibrate systems to false acceptance rates of approximately 0.05%, resulting in false rejection rates between 3.2-4.7% depending on environmental factors and enrollment quality [5]. 3.3. Security Measures and Countermeasures Advanced voice biometric systems incorporate multilayered defenses against fraudulent authentication attempts. Liveness detection forms the primary countermeasure against replay attacks, utilizing specialized algorithms that analyze micro-variations in natural speech patterns including formant transitions, harmonics-to-noise ratios, and spectral flux measurements. These approaches successfully detected 98.2% of replay attacks in Kumar's controlled testing scenarios by identifying unnatural acoustic characteristics that occur when recorded speech is played through speakers [6]. Additional anti-spoofing measures target synthetic speech attacks through analysis of prosodic inconsistencies and phase coherence patterns that typically manifest in artificially generated voices. Evaluation of these techniques across 4 commercially available systems demonstrated detection rates for synthetic speech ranging from 92.7% to 96.5%, with performance correlated to the sophistication of the voice synthesis technology being employed [5]. As voice synthesis capabilities continue to advance, defensive measures have evolved toward multimodal approaches that incorporate behavioral analysis alongside traditional acoustic examination. World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3214 Figure 1 Core Components of Voice Biometric Systems [5, 6] 4. Integration into Customer Experience Journeys 4.1. Strategic Implementation Approaches The integration of voice biometric systems within contact center environments requires careful architectural planning to balance security requirements against operational continuity. According to comprehensive research by Bedi and colleagues, successful implementations follow a phased deployment methodology, with 87% of organizations beginning with limited-scope pilot programs targeting specific customer segments before expanding to broader deployment. Their analysis of implementation processes across financial institutions revealed that hybrid deployment strategies— incorporating both on-premises processing for sensitive operations and cloud-based components for scalability— achieved optimal performance metrics with 99.3% system availability compared to 97.8% for purely cloud-based solutions. The average implementation timeline spans 90-120 days from initial system configuration to full operational capability, with the enrollment phase typically requiring the most significant time investment. Organizations employing dedicated enrollment campaigns successfully registered approximately 62% of their active customer base within the first 30 days of deployment, compared to only 27% adoption for passive enrollment approaches during the same timeframe [7]. Supplementing these findings, Shah's research across diverse industry implementations identified that organizations with formal change management programs achieved 76% higher staff adoption rates and 29% faster time-to-value, underscoring the importance of comprehensive training protocols that encompass both technical operation and customer interaction scripting [8]. 4.2. Authentication Workflow Orchestration The seamless integration of voice biometrics into customer journeys relies on sophisticated decisioning frameworks that govern when and how authentication occurs. Bedi's examination of IVR integration patterns demonstrated that contextual authentication triggers based on transaction type, customer history, and risk assessment algorithms provide superior performance compared to static implementation. Systems employing these dynamic decisioning models experienced 34% lower customer abandonment rates during authentication processes while maintaining equivalent security standards [7]. The orchestration layer typically interfaces with 3-7 separate enterprise systems—including CRM platforms, fraud management tools, and telephony infrastructure—through specialized middleware that coordinates authentication events. According to Shah's technical assessment of integration architectures, RESTful API implementations demonstrate significantly lower latency (averaging 320 ms) compared to SOAP-based approaches (averaging 470 ms) when executing verification operations. The resulting customer experience improvements manifest in measurable metrics, with voice biometric authentication reducing average handling time by 35-45 seconds compared to knowledge-based methods across the 12 contact center implementations evaluated in their study [8]. World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3215 4.3. Omnichannel Considerations and Consistency As organizations expand customer engagement across diverse interaction channels, maintaining consistent authentication experiences presents significant technical challenges. Bedi's research documents the complexities of cross-channel biometric deployment, with audio quality variations significantly impacting performance across different communication modalities. Their comparative analysis demonstrated that narrowband telephony channels (300-3400 Hz) achieved equal error rates (EERs) of 4.5% compared to 2.7% for wideband connections (50-7000 Hz), necessitating channel-specific calibration to maintain uniform security thresholds [7]. Building upon these findings, Shah's work identified that successful omnichannel implementations employ adaptive signal processing techniques that automatically detect channel characteristics and apply appropriate compensation algorithms. Their evaluation of mobile application integration revealed that combining voice authentication with device-specific identifiers created a multilayered verification approach that reduced fraudulent authentication attempts by 84% compared to single-factor methods. Organizations implementing these integrated approaches reported net promoter score (NPS) improvements averaging 18 points following deployment, primarily attributed to the 62% reduction in authentication-related customer effort [8]. Figure 2 Voice Biometric Integration in Customer Journey [7, 8] 5. Addressing Technical and Operational Challenges 5.1. Environmental Resilience and Audio Processing Environmental factors present significant challenges for voice biometric implementations in contact center environments. According to Kaur and colleagues' comprehensive analysis of 42 contact center deployments, background noise represents the most significant impediment to authentication accuracy, with noise levels above 60dB increasing false rejection rates by 18-23% compared to controlled environments. Their research revealed that approximately 31% of authentication failures in operational settings stem from environmental interference rather than actual identity mismatches. To counteract these challenges, modern systems employ multi-stage signal enhancement techniques incorporating spectral subtraction and Wiener filtering, which demonstrate capability to improve signal-tonoise ratios by 9-14dB while preserving critical voice features necessary for authentication [9]. These findings align with a longitudinal study of speech recognition performance, which identified that adaptive noise cancellation algorithms employing deep neural network approaches achieved 43% greater noise suppression compared to traditional filtering methods across diverse acoustic environments. Their analysis demonstrated that systems utilizing these advanced techniques-maintained authentication accuracy within 3.5% of baseline performance even when background noise increased to 75dB, representing significant improvement over earlier generation technologies [10]. World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3216 5.2. Psychological Factors and Vocal Variability Voice characteristics exhibit substantial natural variation due to psychological and physiological factors that challenge biometric system performance. Kaur's analysis documented that emotional states significantly impact fundamental frequency (F0), with stress conditions increasing F0 by an average of 26.3% for female speakers and 14.8% for male speakers compared to neutral emotional states. Their controlled experiments with 384 subjects demonstrated that these emotional variations affected multiple acoustic parameters used in authentication algorithms, resulting in verification scores decreasing by an average of 11.7% during simulated high-stress interactions [9]. These findings are reinforced by Krishnan's work, which tracked authentication performance across varied emotional contexts in realworld deployments, finding that systems without emotional adaptation exhibited false rejection rates up to 3.4 times higher during emotionally charged customer interactions such as dispute resolution calls. Their research documented that implementation incorporating contextual emotional detection algorithms achieved 37% reduction in false rejections during these challenging scenarios while maintaining equivalent security standards by dynamically adjusting decision thresholds based on detected emotional states [10]. 5.3. Longitudinal Performance and Aging Effects Voice characteristics undergo gradual transformation through the natural aging process, creating challenges for longterm authentication accuracy. The comparative analysis of voice parameters across five-year intervals documented agerelated changes affecting multiple acoustic features, including decreases in speaking rate (averaging 2.1% annually after age 60) and shifts in formant frequencies (averaging 1.4% annually). These cumulative changes resulted in voice template degradation that increased equal error rates by approximately 0.5-0.8% annually when templates remained static [9]. Krishnan's seven-year longitudinal study of voice recognition systems corroborated these findings while providing deeper insight into mitigation strategies. Their research evaluated various template management approaches across 12,500 users, determining that passive continuous adaptation techniques—which incrementally update templates after each successful authentication—maintained consistent performance with less than 1.2% degradation over the study period compared to 8.7% degradation for static template approaches. This research further established that hybrid template models employing both foundational templates and incremental updates achieved optimal balance between stability and adaptability, with 94% of successful authentications requiring only a single verification attempt even after multiple years of enrollment [10]. Table 2 Environmental Factors Affecting Voice Biometric Performance [9, 10] Environmental Factor Impact on False Rejection Rate Mitigation Technique Effectiveness of Mitigation Background Noise (>60dB) 18-23% increase Multi-stage signal enhancement 9-14dB improvement in SNR Channel Bandwidth 1.8% difference (narrowband vs. wideband) Adaptive noise cancellation 43% greater noise suppression Network Latency 7% degradation per 100ms increase Edge computing processing 94% reduction in network dependency Audio Compression 11% increase with high compression Compression-resilient features Maintains accuracy within 3.5% of baseline 6. Privacy, Compliance, and Future Outlook 6.1. Regulatory Frameworks and Enterprise Adaptation The implementation of voice biometric systems necessitates careful navigation of increasingly complex privacy regulations across global jurisdictions. A comprehensive analysis of regulatory compliance in biometric deployments identifies significant operational challenges, with organizations implementing voice authentication reporting an average of 312 person-hours dedicated to compliance assessment during initial implementation phases. Their research across 38 financial institutions reveals that 76% of organizations have established dedicated cross-functional governance committees comprising representatives from legal, security, privacy, and operations departments to manage ongoing compliance requirements. These committees typically review system configurations quarterly, with particular focus on consent management frameworks that must adapt to evolving regulatory standards across operating regions. The technical architecture supporting compliance generally incorporates auditable consent tracking those World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3217 records approximately seven distinct data points for each consent event, including timestamp, authentication method, disclosure version, and interaction channel [11]. This aligns with Newton's findings that effective governance frameworks must establish comprehensive audit trails for both authentication events and administrative activities, with NIST guidelines recommending retention of system access logs for a minimum of two years to support potential investigative requirements. Their assessment of organizational readiness indicates that institutions with formal biometric governance programs experience 83% fewer compliance findings during regulatory examinations compared to those with ad hoc approaches [12]. 6.2. Data Protection and Minimization Strategies Securing voice biometric data requires sophisticated technical controls that balance protection requirements against operational needs. Technical assessment of privacy-preserving approaches highlights the effectiveness of edge computing architectures, which process voice data locally before transmitting only encrypted mathematical templates to centralized storage. Their experimental implementation demonstrated that this approach reduced personally identifiable information exposure by 94% while increasing authentication processing time by only 137 milliseconds. Storage segregation represents another critical protection mechanism, with their research documenting that leading implementations maintain separate encrypted databases for biometric templates and identity information, utilizing tokenization to establish secure linkages without creating single points of vulnerability [11]. These approaches align with Newton's evaluation of NIST biometric standards, which emphasizes the importance of cryptographic protection for biometric data both in transit and at rest. Their analysis identifies AES-256 as the predominant encryption standard employed for template protection, with 72% of evaluated systems implementing additional protections including key rotation schedules averaging 90 days and hardware security modules for cryptographic operations. Data minimization principles manifest through template transformation processes that convert raw voice recordings into mathematical representations using proprietary algorithms, rendering the original biometric data practically unrecoverable even if template databases are compromised [12]. 6.3. Technology Evolution and Ethical Considerations Figure 3 Privacy, Compliance, and Future Outlook for Voice Biometric Systems [11, 12] The voice biometric landscape continues rapid evolution through technological innovation and ethical framework development. Research into emerging approaches highlights the transformative potential of federated learning models that enable algorithm improvement without centralizing sensitive voice data. Their experimental implementation demonstrated equal error rate reductions of 28% through distributed model training while maintaining strict data localization. Additionally, their assessment of continuous authentication approaches—which analyze voice throughout entire customer interactions rather than as discrete events—showed potential for more natural user experiences while simultaneously improving fraud detection by 42% through ongoing behavioral analysis [11]. This technological World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3218 evolution must be guided by robust ethical frameworks, with Newton's analysis of NIST standards emphasizing the importance of bias testing across diverse demographic groups. Their evaluation of commercial systems identified accuracy variations of up to 18% between demographic cohorts, highlighting the critical importance of diverse training data and algorithmic fairness assessments. The research further emphasizes the need for transparency in system capabilities and limitations, with NIST guidelines recommending explicit disclosure of performance characteristics including false match rates, false non-match rates, and enrollment failure rates across representative population samples to enable informed deployment decisions [12]. 7. Conclusion Voice biometrics represents a significant advancement in contact center authentication, offering a sophisticated balance between security imperatives and customer experience priorities. As organizations continue to digitally transform their customer interaction channels, voice authentication provides a compelling alternative to cumbersome knowledgebased verification methods. Despite challenges related to environmental factors and natural voice variations, ongoing technological improvements continue to enhance accuracy and resilience against potential vulnerabilities. The successful implementation of voice biometrics requires careful attention to privacy regulations, transparent consent practices, and secure data management protocols. Looking forward, as artificial intelligence and machine learning capabilities evolve, voice authentication systems will become increasingly sophisticated, potentially expanding beyond security applications into personalization and customer experience enhancement. Organizations that thoughtfully deploy voice biometrics today are not merely addressing immediate security concerns but positioning themselves at the forefront of customer-centric innovation in the contact center space. References [1] Siva Venkatesh Arcot and Mahesh Vaijainthymala Krishnamoorthy, "Voice Biometrics in Contact Center: Revolutionizing Security and Customer Experience," International Journal of Innovative Science and Research Technology, Vol. 10, no. 1, Jan. 2025. https://www.researchgate.net/publication/388400988_Voice_Biometrics_in_Contact_Center_Revolutionizing_ Security_and_Customer_Experience [2] Fouad Cherifi et al., "Performance Evaluation Of Behavioral Biometric Systems," ResearchGate, Jan. 2009. https://www.researchgate.net/publication/257365249_Performance_Evaluation_Of_Behavioral_Biometric_Sys tems [3] Warda Hassan and Nosheen Sabahat, "Towards Secure Identification: A Comparative Analysis of Biometric Authentication Techniques," VFAST Transactions on Software Engineering, Vol. 12, No. 1, March 2024. https://www.researchgate.net/publication/380086939_Towards_Secure_Identification_A_Comparative_Analys is_of_Biometric_Authentication_Techniques [4] Joshua Boluwatife Adelusi, "Voice Biometrics for Authentication: A Comprehensive Exploration," ResearchGate, Jan. 2024. https://www.researchgate.net/publication/387060240_Voice_Biometrics_for_Authentication_A_Comprehensiv e_Exploration [5] Nakshatra Joshi et al., "Voice Biometrics – A New Outlook to a Traditional Problem," International Research Journal of Engineering and Technology (IRJET), Vol. 11, no. 5, May 2024. https://www.irjet.net/archives/V11/i5/IRJET-V11I580.pdf [6] Karuna Grover and Rajesh Mehra, "Biometric Face Anti-Spoofing and Context-Based Detection Techniques: A Review," JETIR, Vol. 6, no. 4, April 2019. https://www.jetir.org/papers/JETIR1904R22.pdf [7] Nilu Singh et al., "Voice Biometric: A Technology for Voice Based Authentication," Advanced Science, Engineering and Medicine, Vol. 10, no. 7, July 2018. https://www.researchgate.net/publication/324031666_Voice_Biometric_A_Technology_for_Voice_Based_Auth entication [8] Amjad Hassan Khan M. K. and P. S. Aithal, "Identification of Customer Through Voice Biometric System in Call Centres," I.J. Intelligent Systems and Applications, 8 Oct. 2024. https://www.mecs-press.org/ijisa/ijisa-v16n5/IJISA-V16-N5-6.pdf World Journal of Advanced Research and Reviews, 2025, 26(02), 3211-3219 3219 [9] Arne De Keyser et al., "Opportunities and challenges of using biometrics for business: Developing a research agenda," Journal of Business Research, Vol. 136, Nov. 2021. https://www.sciencedirect.com/science/article/abs/pii/S0148296321005014 [10] Ingo Siegert et al., "Recognition Performance of Selected Speech Recognition APIs - A Longitudinal Study," ResearchGate, Oct. 2020. https://www.researchgate.net/publication/345307161_Recognition_Performance_of_Selected_Speech_Recogni tion_APIs_-_A_Longitudinal_Study [11] Zaid Sh. Alattar et al., "Privacy-preserving hands-free voice authentication leveraging edge technology," ResearchGate, Dec. 2022. https://www.researchgate.net/publication/366328305_Privacy-preserving_handsfree_voice_authentication_leveraging_edge_technology [12] NIST, "Standards for Biometric Technologies," 19 June 2013. https://www.nist.gov/speechtestimony/standards-biometric-technologie