Quantum-Trained Convolutional Neural Network for Deepfake Audio Detection

This paper introduces QT-CNN, a hybrid Quantum-Trained Convolutional Neural Network architecture designed to enhance deepfake audio detection through quantum machine learning techniques. Key features include:

• Quantum-to-classical parameter mapping that enables up to 70% reduction in trainable parameters without sacrificing detection accuracy.
• A hybrid architecture combining Quantum Neural Networks (QNNs) with classical CNN layers for efficient and expressive model training.
• Comprehensive audio preprocessing pipeline, including feature extraction, label encoding, and sequential dataset construction.

Experimental evaluations show that QT-CNN matches the accuracy of classical CNNs while offering improved computational efficiency, making it highly suitable for resource-constrained environments in media forensics and audio security.

Recommended citation: Lin, Chu-Hsuan Abraham; Liu, Chen-Yu; Chen, Samuel Yen-Chi; & Chen, Kuan-Cheng. (2025). "Quantum-Trained Convolutional Neural Network for Deepfake Audio Detection." Proceedings of ICASSP 2025 – IEEE International Conference on Acoustics, Speech, and Signal Processing.
Download Paper