Waveform and Mel-Frequency Cepstral Coefficients (MFCC) approach for Deepfake Audio Detection

• Published in: 2025 IEEE International Conference on Emerging Technologies and Applications (MPSec ICETA) pp. 1 - 6
• Main Authors: Singh, Sunny, Bharadwaj, Nitesh K
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.02.2025
• Subjects: Accuracy; Analytical models; Authentication; CNN+LSTM; Deep4SNet model; Deepfake audio; Deepfakes; Hybrid; Identity theft; Long short term memory; Prevention and mitigation; Privacy; Safety; Security; Security and privacy; WaveNet
• DOI: 10.1109/MPSecICETA64837.2025.11118516

The ability for deepfake audio technology to manipulate voices represents a threat to the very integrity of audio authentication systems liabilities for scams, identity theft, and erosion of trust in digital communication. As these convincing audio clips of the voice of individuals and organisations begin to emerge, they become subject to those higher risks of anything from financial fraud to reputational harm. In our assessment of these findings, we have presented by way of admonition to choose models for audio authentication that epitomise safety and security in a time that is becoming enormously influenced by advancement in technology. This shows a covenant in which a combination of Deep4SNet, WaveNet and the hybrid CNN + LSTM architecture is employed to testify for manipulated audio cases. Thus, we found varying accuracies, whereby the combinations of Deep4SNet showed a higher perceived accuracy at 83 \%-a monumental call to action for researchers and practitioners to build efficient defenses against deepfake threats elsewhere.