MA-DLE: Speech-based Automatic Depression Level Estimation via Memory Augmentation

In three linesMA-DLE introduces a memory augmentation method for automatic depression level estimation from speech. The system enhances GRU-extracted features using a selective memory bank (historical temporal features + dynamic memory features based on variability) and a Hierarchical Attention Fusion module. Evaluated on DAIC-WOZ and E-DAIC datasets, it achieves state-of-the-art performance.

Summary generated by Claude — human-verified