Pragmatics of Synthetic Deception: Speech Act Analysis and Psycholinguistic Manipulation in Deepfake Voice Scam Messages in Indonesia

Abstract

Voice-cloned deepfake scams have become a fast-growing vector of financial fraud in Indonesia, with reported losses reaching the trillions of rupiah and a documented surge exceeding 1,500% in incident volume between 2022 and 2023. Forensic and cybersecurity literature has mapped the technical signatures of synthetic audio, but comparatively little work has examined how these synthetic voices linguistically perform deception: what speech acts they enact and how their prosodic design exploits listener cognition. This study addresses that gap by analyzing eight Indonesian-language deepfake scam voice samples, reconstructed from publicly reported cases and verified incident patterns, through Searle's (1976) taxonomy of illocutionary acts combined with discourse-psycholinguistic analysis of pause structure and pitch (F0) contour using Praat-based acoustic modeling. Results show that directive acts dominate the corpus (36% of 95 coded units), typically preceded by assertive acts that construct false authority, and that scam discourse follows a recurring five-phase structure: authority framing, crisis-trigger diction, a manipulative pre-directive pause, directive command, and closing reassurance, a sequence absent from ordinary conversational speech. The pre-directive pause (M = 980 ms) runs roughly three times longer than the conversational baseline (M = 340 ms) and functions as a cognitively engineered interval that suppresses critical evaluation before the financial demand is issued. These findings extend speech act theory into synthetic media contexts and show that deepfake persuasion is a deliberately engineered pragmatic structure, not just a visual or technical artifact. The study contributes a replicable analytic framework for forensic linguists, digital literacy educators, and policymakers seeking to counter AI-generated voice fraud.