EUVIMEDEUVIMED entdecken
Crossref

Fusing Self-Supervised Speech Representations with MFCC Stability for Robust Deepfake Audio Detection

Sadineni Havesa, Golla Susanth Paul, S. Jagadeesan

International Journal of Innovative Science and Research Technology · 09.04.2026 · 3636

DOI: 10.38124/ijisrt/26mar2080 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Innovative Science and Research Technology, online erschienen am 09.04.2026. Verfasst von Sadineni Havesa und 2 weiteren Autor:innen. Die Arbeit nennt 17 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recent advancements in neural speech synthesis have enabled the creation of extremely realistic deepfakes that sound like actual human voices. Even though these technologies have some useful purposes, they also pose serious risks in terms of voice impersonation, misinformation, and financial scams. The detection of fake speech is an emerging research concern in speech forensics and cybersecurity. This paper proposes a dual-branch deep learning framework for effective deepfake audio detection by combining self-supervised speech representations with handcrafted acoustic stability features. The first branch is responsible for the extraction of semantic speech embeddings using the pre-trained model WavLM, which contains contextual and phonetic information from the speech signal. The second branch is responsible for the extraction of Mel-Frequency Cepstral Coefficient (MFCC) stability features and the application of the Temporal Convolutional Network (TCN) model. The features from both branches are combined using a multilayer perceptron classifier to decide if an audio sample is real or fake. Experiments on the Fake-or-Real (FoR) dataset show that the proposed fusion approach enhances detection performance compared to models using a single feature. The results suggest that merging deep contextual embeddings with handcrafted stability features offers better resilience against modern deepfake audio generation methods.

Zitieren

APA

Havesa, S., Paul, G. S., & Jagadeesan, S. (2026). Fusing Self-Supervised Speech Representations with MFCC Stability for Robust Deepfake Audio Detection. International Journal of Innovative Science and Research Technology, 3636. https://doi.org/10.38124/ijisrt/26mar2080

Vancouver

Havesa S, Paul GS, Jagadeesan S. Fusing Self-Supervised Speech Representations with MFCC Stability for Robust Deepfake Audio Detection. International Journal of Innovative Science and Research Technology. 2026:3636. doi:10.38124/ijisrt/26mar2080

BibTeX
@article{Havesa2026,
  author = {Havesa, Sadineni and Paul, Golla Susanth and Jagadeesan, S.},
  title = {Fusing Self-Supervised Speech Representations with MFCC Stability for Robust Deepfake Audio Detection},
  journal = {International Journal of Innovative Science and Research Technology},
  year = {2026},
  pages = {3636},
  doi = {10.38124/ijisrt/26mar2080}
}
Kostenlos starten →