Publication

Abstract : The capacity to identify real audio recordings from their modified counterparts is essential in the age of sophisticated digital manipulation for maintaining security and trust in a vari- ety of applications, from media forensics to voice authentication systems. This research aims to create a deep learning model that can distinguish between authentic and altered audio files, with an emphasis on identifying audio deepfakes. The study uses Mel spectrogram representations and data augmentation techniques to effectively extract features from the ASV spoof 2019 dataset and train models. Convolutional neural networks (CNNs) comprising a number of layers, including convolutional, pooling, batch normalization, ReLU activation, dropout, global average pooling, and a dense classification layer are used as the foundation of the design. The Adam optimizer is used to optimize the model once it has been trained using binary cross-entropy loss, and a variety of metrics, such as accuracy, F1 score, ROC curve, and AUC, are used to track its performance. By making it easier to identify audio deepfakes, this project will ultimately increase the security and integrity of audio data in the digital world.