Acoustic-Based Detection of AI-Generated Vishing Attacks in Telecom Environments
A Machine Learning Framework for Voice Feature Analysis and Fraudulent Call Identification
📄
Abstract
Abstract:
patterns embedded in AI-generated speech. The CNN-based system outperforms standalone SVM with a uniform score of 0.88, while the CNN-based architecture is more competitive in terms of feature representation, by learning hierarchical acoustic representations directly from the raw spectral input with less reliance on hand-crafted feature engineering. With Explainability based on LIME, further the rationales of decisions are auditable, which are applicable for anti-spoofing guidelines by TRAI and STIR/SHAKEN standard and keep the framework technically sound and legally responsible.
🏷️
Keywords
📖
Full Article
🖼️
Figures & Tables
📚
References
⚖️
Ethics Declarations
Ethics Approval and Consent to Participate
Users can freely download, share, and reuse the article for any purpose (even commercially), as long as they give proper credit to the original author.
Competing Interests
Users can freely download and share the article, but they are strictly prohibited from using it or selling it for commercial profit.
Consent for Publication
Users can only read and view the article for personal use; they cannot copy, redistribute, or reuse any part of it without explicit permission from your journal.
📋
Editorial Lifecycle
Audit TrailComplete editorial process — cryptographically verified
Editorial Lifecycle
Audit TrailEditorial Process Audit Trail
NFSU-2026-V5-I1-J-2026-0116-CHAIN
✓ Verified · 0 blocks · Genesis 0x9f2bNo History Found
This manuscript has no status history yet.
🔐
Verification & Trust
CompliantCOPE · ICMJE · FAIR · ORCID · CLOCKSS
Verification & Trust
CompliantArticle Integrity Verified
All editorial chain blocks cryptographically verified · Compliant with COPE, ICMJE, and FAIR standards