Comparative Analysis Of ML/DL Models for ‎Voice Spoofing Detection

Abstract

Spoofing of voice is an alarming threat to voice-security systems, especially in high-stakes areas like ‎banking, smart home gadgets, customer support, and virtual assistants. In this paper, the authors submit an ‎inclusive comparative review of spoof detection methods with a description of how they evolved from ‎classical machine learning to the latest deep learning and transformer-based models. The performance of ‎different models is measured across benchmark datasets with common metrics like Equal Error Rate ‎‎(EER) and tandem Detection Cost Function (t-DCF). Interestingly, the increase in the effectiveness of ‎Transformer architectures for spoofed audio detection is emphasized. The ultimate goal of the present ‎research is to assist both professionals and academics in selecting and developing dependable and ‎safe voice authentication systems‎.