VoiceSentry: Hybrid Neural Network Approach for Deep Fake Audio Recognition

Abstract

The rapid rise of synthetic media has created serious concerns regarding the authenticity of audio content. Detecting deep fake audio has become increasingly important, especially in areas such as digital security and media verification. This project presents a webbased system that uses deep learning techniques to classify audio files as either real or fake. The model combines Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to capture both feature patterns and time-based dependencies in audio signals. The system allows users to upload datasets, process audio files, and train the model through a simple interface. During experimentation, the dataset is divided into training and testing portions to evaluate performance effectively. The trained model achieves an accuracy of around 95%, along with strong precision and recall values. Visual tools such as confusion matrix and training graphs are used to better understand the model’s behavior. In addition, users can upload new audio samples and get instant predictions. The system performs well in identifying manipulated audio, with only a small number of incorrect predictions. Overall, the project provides a practical and efficient solution for deep fake audio detection and can be useful in real-world applications. Keywords—Deep fake audio detection, audio classification, deep learning, convolutional neural network (CNN), long short-term memory (LSTM), hybrid model, audio preprocessing, feature extraction, training-testing split, confusion matrix, performance evaluation, accuracy, precision, recall, F1-score, real-time detection, web-based application, audio forensics, media verification, user interface