International Journal of Innovative Research in Computer and Communication Engineering

The development of deepfake technology Shows a serious threat to the integrity of voice communications. with potential misuse in identity theft, misinformation, and fraud. The main goal of the project is to create a powerful detection system that can distinguish real sound using the audio signal to increase the security of audio files. To achieve this, we propose a novel approach that integrates Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. The CNN component extracts spatial features from audio spectrograms, identifying unique patterns that differentiate real and fake audio. These extracted features are then processed by the Bi-LSTM, which captures temporal dependencies and context, further improving detection accuracy. The proposed model was tested on the “Lj speech dataset” and “wavefake dataset” , demonstrating superior performance over conventional methods with improved generalization across various spoofing techniques. This approach is particularly applicable in enhancing security in voice authentication systems, digital forensics, and other domains requiring reliable detection of audio manipulations.