Multi-feature deep supervised voiceprint adversarial network for depression recognition from speech

Yuchen Pan, Yuanyuan Shang, Wei Wang, Zhuhong Shao, Zhuojin Han, Tie Liu, Guodong Guo, Hui Ding

Abstract: Highlights•The MFDS-VAN uses various features to robustly capture depression information.•EN module uses attention-equipped sub-modules feature extraction and fusing.•DSR algorithm improves convergence, accuracy, and robustness through optimization.•VAN boosts vocal features and reducing the impact of individual voiceprints.