EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Explainable multimodal deepfake detection across media platforms: A cross-modal transformer framework for trust

Hari Krishna Silamanthula, Yogesh Babu Thunga, Haneef Bapana, Suresh Babu Malakari

International Journal of Applied Resilience and Sustainability · 07.06.2026 · 2 (3) · 289-305

DOI: 10.70593/deepsci.0203011 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Applied Resilience and Sustainability, online erschienen am 07.06.2026. Verfasst von Hari Krishna Silamanthula und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The synthetic media is more rapidly spread than people could check it. Deepfakes are fake faces, cloned voices, and manipulated text. They traverse through numerous media outlets within a few seconds. A majority of the detectors operate separately. They are also black-box. Users view a label and no rationale. This breaks trust. In single-channel forgeries, single-channel tools do not work when the forgery is clean in a given channel and weak in a different one. This work presents and tests a cross-modal transformer model of explainable multimodal deepfake detection. The objective is to combine visual, audio, and textual stimulations in addition to revealing reasons behind every choice. We used method of simulation and modeling. We construct an artificial architecture and a synthetic data generator that simulates platform-level distortion. We refer to the fusion core as cross-modal attention. We add a head of explanation which generates the scores of modality attribute and the calibration of trust. We carry out regulated simulations in video, audio and social streams of text. The simulated-run results have demonstrated that cross-modal fusion is significantly elevating the detectability to levels which are well beyond the individual-modality benchmarks. The framework also maintains the calibration error minimal. Attention attribution goes in line with the modality that conveys the forgery. The explainability layer increases an index of human trust modeled without decreasing the accuracy. These patterns are discussed in comparison to earlier work that was unimodal only and late-fusion. It illustrates why transformer-based cross-modal reasoning can be beneficial to media forensics and misinformation detection.

Zitieren

APA

Silamanthula, H. K., Thunga, Y. B., Bapana, H., & Malakari, S. B. (2026). Explainable multimodal deepfake detection across media platforms: A cross-modal transformer framework for trust. International Journal of Applied Resilience and Sustainability, 2(3), 289-305. https://doi.org/10.70593/deepsci.0203011

Vancouver

Silamanthula HK, Thunga YB, Bapana H, Malakari SB. Explainable multimodal deepfake detection across media platforms: A cross-modal transformer framework for trust. International Journal of Applied Resilience and Sustainability. 2026;2(3):289-305. doi:10.70593/deepsci.0203011

BibTeX
@article{Silamanthula2026,
  author = {Silamanthula, Hari Krishna and Thunga, Yogesh Babu and Bapana, Haneef and Malakari, Suresh Babu},
  title = {Explainable multimodal deepfake detection across media platforms: A cross-modal transformer framework for trust},
  journal = {International Journal of Applied Resilience and Sustainability},
  year = {2026},
  volume = {2},
  number = {3},
  pages = {289--305},
  doi = {10.70593/deepsci.0203011}
}
Kostenlos starten →