Data-Driven Techniques for Speech and Multimodal Deepfake Detection

Read full text Summarise with VIRAT
Authors: Davide Salvi
Cited by: 0
References: 21

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Recent advancements in deep learning and generative models have simplified the creation and manipulation of synthetic media. Today, even inexperienced users can produce highly realistic content with minimal effort. While these technologies offer exciting opportunities, they also pose serious risks. When misused, they can facilitate fraud, blackmail, and the spread of disinformation. An example of this phenomenon is deepfakes, synthetic multimedia content generated through deep learning techniques that depict individuals in actions and behaviors that do not belong to them. Using only a few images or an audio recording of a target victim, an attacker can utilize deepfake technology to produce synthetic data that impersonates the victim and discredits their reputation. Detecting such content is essential to prevent misuse. This chapter addresses the problem of deepfake detection, beginning with a monomodal focus on synthetic speech and then extending the analysis to audio-video multimodal deepfakes. We propose multiple detection methods and discuss broader solutions to related challenges. We view this work as a foundational but meaningful step forward in multimedia forensics. While the results are encouraging, the landscape is evolving rapidly, with emerging threats demanding continuous innovation. We believe our findings can support future research and help strengthen defenses against synthetic media misuse.

Fields of Study & Subjects

Digital Media Forensic Detection Speech Recognition and Synthesis Misinformation and Its Impacts

Publication details

Journal / source
SpringerBriefs in applied sciences and technology
Pages
107–116
Publication date
26 March 2026
Language
EN
Document type
Book chapter