Detection of Unseen Voice-Cloning Attacks

The rapid advancement of artificial intelligence-based speech synthesis and voice-cloning technologies has made it increasingly difficult to distinguish genuine human speech from artificially generated or cloned voices. Although existing detection techniques can perform strongly on known synthetic speech, their reliability against previously unseen voice-cloning attacks remains a significant research challenge. This paper investigates the problem of detecting unseen voice-cloning attacks with emphasis on generalization, robustness, and evaluation under changing attack conditions. The review covers acoustic and spectral analysis, anti-spoofing systems, one-class learning, zero-shot detection, domain generalization, adversarial attacks, and real-time detection. Based on the identified gap, the proposed research design uses a feature-based one-class detection baseline in which genuine speech is modeled and evaluated against both known and unseen cloned speech. The methodology includes audio normalization, extraction of spectral and cepstral characteristics, one-class classification, cross-generator evaluation, and robustness testing under compression and signal-processing variations. The research design is intended to measure the generalization gap using accuracy, precision, recall, F1-score, and Equal Error Rate. Because no experimental audio dataset or executed model outputs were supplied with the source material, this paper does not present invented numerical experimental results; instead, it provides a complete evaluation protocol and a results structure that can be populated after execution. The literature synthesis indicates that performance on known attacks can overestimate real-world reliability and that unseen generators, post-processing, and adversarial modifications can reduce detection effectiveness. The study therefore emphasizes generator-independent characteristics, explicit unseen-attack evaluation, and open-world robustness as essential requirements for dependable voice-cloning detection in cybersecurity, authentication, telecommunications, and digital forensics.

Authors

Institutions

Publication Details

Journal
International Journal for Research in Applied Science and Engineering Technology
Published
2026-11-16
DOI
https://doi.org/10.22214/ijraset.2026.85696
Primary Topic
Speech Recognition and Synthesis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Detection of Unseen Voice-Cloning Attacks

Dhanashri Abhang, Koushik Jaiswal, Chandan Sharma
5 links
International Journal for Research in Applied Science and Engineering Technology
Speech Recognition and Synthesis
article

Detection of Unseen Voice-Cloning Attacks

Dhanashri Abhang, Koushik Jaiswal, Chandan Sharma
article en

Abstract

The rapid advancement of artificial intelligence-based speech synthesis and voice-cloning technologies has made it increasingly difficult to distinguish genuine human speech from artificially generated or cloned voices. Although existing detection techniques can perform strongly on known synthetic speech, their reliability against previously unseen voice-cloning attacks remains a significant research challenge. This paper investigates the problem of detecting unseen voice-cloning attacks with emphasis on generalization, robustness, and evaluation under changing attack conditions. The review covers acoustic and spectral analysis, anti-spoofing systems, one-class learning, zero-shot detection, domain generalization, adversarial attacks, and real-time detection. Based on the identified gap, the proposed research design uses a feature-based one-class detection baseline in which genuine speech is modeled and evaluated against both known and unseen cloned speech. The methodology includes audio normalization, extraction of spectral and cepstral characteristics, one-class classification, cross-generator evaluation, and robustness testing under compression and signal-processing variations. The research design is intended to measure the generalization gap using accuracy, precision, recall, F1-score, and Equal Error Rate. Because no experimental audio dataset or executed model outputs were supplied with the source material, this paper does not present invented numerical experimental results; instead, it provides a complete evaluation protocol and a results structure that can be populated after execution. The literature synthesis indicates that performance on known attacks can overestimate real-world reliability and that unseen generators, post-processing, and adversarial modifications can reduce detection effectiveness. The study therefore emphasizes generator-independent characteristics, explicit unseen-attack evaluation, and open-world robustness as essential requirements for dependable voice-cloning detection in cybersecurity, authentication, telecommunications, and digital forensics.

International Journal for Research in Applied Science and Engineering TechnologyVol. 14(11)
JSPM University (IN)
Speech Recognition and Synthesis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon