Detecting Deepfake Audio Using Spectrogram-Based Machine Learning Approaches

This project focuses on detecting deepfake audio using spectrogram-based machine learning approaches, applied to the classification of real and fake audio samples. The dataset consists of two classes: "Real" (0) and "Fake" (1), representing authentic and deepfake audio respectively. Various machine learning models, including Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) networks, were employed for classification tasks. Spectrograms of audio signals were used as input features to capture the temporal and frequency characteristics of the audio data. A user-friendly web application was developed using Flask, HTML, CSS, and JavaScript, allowing users to upload audio samples for real-time deepfake detection. This solution contributes to the advancement of deepfake audio detection, enhancing security and media integrity across various applications.

Keywords: Deepfake Audio Detection, Spectrogram, Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Flask, Audio Classification, Fake Audio Detection, Predictive Maintenance.