Emotional Vietnamese Speech-Based Depression Diagnosis Using Dynamic Attention Mechanism
The study developed a Dynamic Attention GRU Network to classify emotions in Vietnamese speech signals to diagnose depression tendency. It uses a dual-stream approach combining raw waveform and MFCC features, enhanced by a novel Dynamic-CBAM block. The model achieved high performance (F1-score of 0.87) on the VNEMOS dataset, demonstrating its effectiveness in extracting emotion-relevant features for mental health diagnosis.