Acoustic signature analysis for distinguishing human vs. synthetic voices in vishing attacks

Abstract

Cybercrimes targeting mobile devices are on the rise, with vishing and smishing attacks being particularly prevalent. These attacks exploit social engineering techniques to manipulate individuals into divulging personal information or engaging in unintended actions. To counter this evolving threat landscape, this research proposes a pioneering methodology rooted in voice feature analysis. By distinguishing between human and robotic voices, this approach aims to discern legitimate calls from potential scams, thereby mitigating the associated financial losses and reputational damage. The research delves into the intricacies of voice feature analysis, leveraging natural language processing (NLP) and machine learning (ML) techniques to extract and analyze audio attributes such as pitch, volume, and temporal patterns. The ultimate objective is to create a binary classification model that accurately differentiates between human voice calls and robocalls, incorporating a comprehensive dataset comprising actual call recordings and synthesized scenarios. This research advances beyond conventional practices by championing a holistic analysis of both human and robocalls, contrary to the prevalent assumption of robocalls exclusively constituting scams. The application of various audio features, coupled with nuanced weightage allocation, enhances the model’s discernment capabilities. The resultant binary classifier is an exemplar of the innovative fusion of technology and human expertise. In conclusion, this research introduces a novel dimension to the combat against vishing and smishing attacks, with a robust voice feature analysis methodology capable of accurately identifying human and robotic voices. By effectively distinguishing legitimate calls from potential threats, this approach presents a promising avenue for safeguarding individuals and organizations against the far-reaching consequences of cybercrimes. The comprehensive analysis, validation, and insights presented in this paper contribute significantly to the field of cybersecurity and voicebased communication analysis.