The field of signal processing and security is moving towards innovative solutions to address the increasing challenges in radio frequency spectrum utilization, speech synthesis, and audio authentication. Researchers are exploring complex-valued neural networks, adversarial training, and self-supervised learning models to improve the accuracy and robustness of signal classification and speech detection systems. Noteworthy papers include: CV-MuSeNet, which achieves state-of-the-art performance in wideband spectrum sensing with a complex-valued multi-signal segmentation network; ASRJam, which proposes a proactive defense framework against automated phone scams using adversarial perturbations; and Pushing the Performance of Synthetic Speech Detection with Kolmogorov-Arnold Networks and Self-Supervised Learning Models, which improves synthetic speech detection performance by 60.55% relatively on LA and DF sets.
Advances in Signal Processing and Security
Sources
Attention-based Adversarial Robust Distillation in Radio Signal Classifications for Low-Power IoT Devices
A Neural Rejection System Against Universal Adversarial Perturbations in Radio Signal Classification