Abstract

Cloned voices are now cheap to make and easy to misuse in phone scams that imitate a relative or a colleague. This review paper examines what the research literature offers for defending ordinary users against such calls and uses it to propose Voice-Clone Guardian, a mobile application that runs entirely on the phone. The application combines two checks that are usually studied separately: an anti-spoofing model that looks for signs of synthetic speech, and a speaker-verification model that compares the voice with a profile enrolled for a family member. The paper reviews benchmark, model and survey studies, justifies the choice of a lightweight graph-attention countermeasure (AASIST-L class) and an ECAPA-TDNN speaker encoder, and describes the audio front-end, enrolment procedure, scoring pipeline and decision rules, with a diagram for each. The literature also shows clear limits: detectors trained on benchmark data degrade sharply on real-world audio, short clips give weak evidence, and mobile operating systems block direct access to call audio. The work is at the review and design stage; no prototype has been built and no performance figures are claimed for the proposed system. An evaluation plan is given for the next phase.

Keywords

How to Cite This Article

Rights