Fake Voice Detection: A Hybrid CNN-LSTM Based Deep Learning Approach

This study focuses on developing and evaluating a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) based deep learning model for detecting fake voice recordings. The proposed model addresses the critical issue of artificial intelligence-generated speech mimicking human voices, which can potentially be used for malicious purposes, thereby endangering individuals' privacy and safety. A comprehensive dataset comprising 5,889 real and 5,889 fake voice samples was utilized for this research. The dataset underwent rigorous preprocessing, including segmentation into fixed-length windows and normalization. The hybrid CNN-LSTM model was then trained and validated systematically involving exploratory data analysis and extensive hyperparameter tuning. The experimental results demonstrated that the proposed model achieved an accuracy of 99.2%, an F1 score of 99.2%, a recall of 99.4%, and a precision of 99.0%, indicating its robust performance in distinguishing between real and fake voices. The findings underscore the potential of the hybrid CNN-LSTM model as a powerful tool for safeguarding digital communications against the growing threat of fake voices. © 2024 IEEE.