A comprehensive review on detecting partially spoofed audio in speaker verification
Dinithi Gunawardena, Saadh Jawwadh · 2025
The advancement of deepfake technology, especially in the audio domain, has introduced significant challenges to the integrity of voice-based systems. While much research has focused on detecting fully spoofed audio, partially spoofed audio, where only segments of speech are synthetically manipulated, remains a growing and underexplored threat to automatic speaker verification (ASV) systems. This review offers an in-depth exploration of the existing state of audio deepfake detection, emphasising the emerging challenge of partially spoofed audio. It categorises major types of audio deepfakes, summarises benchmark challenges, and reviews available datasets. The paper also explores traditional machine learning (ML), deep learning (DL), and hybrid approaches. The review concludes with future directions aimed at developing robust, explainable, and real-time detection systems capable of addressing this evolving threat.