AI-Based Deepfake Voice Detection System

This research introduced a real-time solution for identifying AI- generated deepfake voices, tackling an escalating issue in cybersecurity, communication, and fraud mitigation. By integrating MFCC, spectral, and prosodic features with both traditional machine learning techniques (Random Forest, SVM, Logistic Regression) and deep learning approaches (CNN, LSTM, CNN-LSTM), the system demonstrated strong performance on recognized benchmark datasets. The combination of a React-based web interface for software notifications and an Arduino-based physical alert mechanism made the tool user-friendly and offered immediate feedback to users. Although there are limitations, such as vulnerability to new synthesis methods, real-time performance challenges on edge devices, and the possibility of false positives or negatives, the system lays a solid groundwork for effective deepfake voice detection. Future enhancements in multilingual capabilities, lightweight deployment on edge devices, continuous learning, and resilience against adversarial attacks will improve its robustness and usability, making such systems critical for safeguarding individuals, organizations, and society against the increasing dangers of audio-related fraud and misinformation

[1]  J. Yamagishi, M. Todisco, and N. Evans, “ASVspoof 2019: A large-scale public database for synthetic speech spoofing,” IEEE Transactions on Biometrics, Behavior, and Identity Science, vol. 3, no. 2, pp. 216–227, 2021.   [2]     A. Nautsch, X. Wang, N. Evans, T. Kinnunen, et al., “ASVspoof 2021: Automatic speaker verification spoofing and countermeasures challenge evaluation plan,” arXiv preprint arXiv:2109.00537, 2021.   [3]   J. Patino, M. Todisco, and N. Evans, “Speaker verification spoofing countermeasures: A survey,” Speech Communication, vol. 141, pp. 1–28, 2022.   [4]   A. K. Tiwari and A. Singh, “Deep learning approaches for detecting voice spoofing attacks: A review,” Journal of Information Security and Applications, vol. 70, pp. 103–126, 2023.   [5]  R. K. Das, J. Yang, T. Kinnunen, Z. Wu, et al., “Long-term spoofing countermeasures: A new challenge for automatic speaker verification,” Computer Speech & Language, vol. 77, p. 101418, 2023.   [6]   H. Wu, Y. Jia, and J. Tao, “WaveFake: A benchmark for audio deepfake detection,” Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022.   [7]    S. Khalid, S. Javed, and M. Hussain, “A comprehensive survey on deepfake detection techniques in audio, image, and video domains,” ACM Computing Surveys, vol. 56, no. 4, pp. 1–38, 2024.   [8]   M. Alzantot, Z. Wang, and M. Srivastava, “Deep residual neural networks for speech spoofing detection,” Proc. Interspeech, pp. 1020–1024, 2019.   [9]   S. R. J. Ramakrishnan and R. V. Babu, “Speech deepfake detection using spectrogram and convolutional recurrent neural networks,” IEEE International Conference on Pattern Recognition (ICPR), 2022.   [10]   P. Korshunov and S. Marcel, “Vulnerability assessment and detection of audio deepfake attacks,” IEEE International Conference on Biometrics (ICB), pp. 1–8, 2018.