The proliferation of audio deepfakes poses significant risks to modern society, ranging from misinformation to secu- rity threats. As deepfake technology continues to advance, the need for robust detection mech- anisms becomes critical. This thesis introduces a novel approach for detecting audio deepfakes utilizing implicit neural representations. By leveraging these representations, we aim to capture and analyze the unique patterns inherent in audio, particularly focusing on the segments of si- lence. The performances of the model are close to the state-of-the-art, proving the effectiveness of this methodology. To assess the ability of humans and algorithms to detect audio deepfakes, perceptive tests in a controlled environment were performed, and compared against a baseline model. The results show that human testers perform poorly, in comparison to the baseline model. These insights might be helpful for future research.

The proliferation of audio deepfakes poses significant risks to modern society, ranging from misinformation to secu- rity threats. As deepfake technology continues to advance, the need for robust detection mech- anisms becomes critical. This thesis introduces a novel approach for detecting audio deepfakes utilizing implicit neural representations. By leveraging these representations, we aim to capture and analyze the unique patterns inherent in audio, particularly focusing on the segments of si- lence. The performances of the model are close to the state-of-the-art, proving the effectiveness of this methodology. To assess the ability of humans and algorithms to detect audio deepfakes, perceptive tests in a controlled environment were performed, and compared against a baseline model. The results show that human testers perform poorly, in comparison to the baseline model. These insights might be helpful for future research.

Audio Deepfake Detection based on Implicit Neural Representations

SIMONATO, NICCOLÒ
2023/2024

Abstract

The proliferation of audio deepfakes poses significant risks to modern society, ranging from misinformation to secu- rity threats. As deepfake technology continues to advance, the need for robust detection mech- anisms becomes critical. This thesis introduces a novel approach for detecting audio deepfakes utilizing implicit neural representations. By leveraging these representations, we aim to capture and analyze the unique patterns inherent in audio, particularly focusing on the segments of si- lence. The performances of the model are close to the state-of-the-art, proving the effectiveness of this methodology. To assess the ability of humans and algorithms to detect audio deepfakes, perceptive tests in a controlled environment were performed, and compared against a baseline model. The results show that human testers perform poorly, in comparison to the baseline model. These insights might be helpful for future research.
2023
Audio Deepfake Detection based on Implicit Neural Representations
The proliferation of audio deepfakes poses significant risks to modern society, ranging from misinformation to secu- rity threats. As deepfake technology continues to advance, the need for robust detection mech- anisms becomes critical. This thesis introduces a novel approach for detecting audio deepfakes utilizing implicit neural representations. By leveraging these representations, we aim to capture and analyze the unique patterns inherent in audio, particularly focusing on the segments of si- lence. The performances of the model are close to the state-of-the-art, proving the effectiveness of this methodology. To assess the ability of humans and algorithms to detect audio deepfakes, perceptive tests in a controlled environment were performed, and compared against a baseline model. The results show that human testers perform poorly, in comparison to the baseline model. These insights might be helpful for future research.
Deepfake Detection
Deepfake
Implicit
Neural
Representations
File in questo prodotto:
File Dimensione Formato  
Cybersecurity_MsC_Thesis_on_Audio_DeepFake_Detection___Simonato_PDFA.pdf

accesso aperto

Dimensione 9.88 MB
Formato Adobe PDF
9.88 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/80281