This project provides the code of using emotionally rich speech segments for depression prediction.
We think that emotionally rich speech segments provide more salient cues that can distinguish between depressed and non-depressed individuals. In emotionally rich segments, such as those with high or low valence, differences in emotional expression between depressed and non-depressed individuals may be amplified, making depression easier to detect.
To use this code, you first need to prepare your data, including extracting arousal, valence, and dominance scores, as well as PDEM embeddings, using the public dimensional emotion model. The scripts 1_extract_AVD_scores_from_PDEM.py and 2_extract_embedding_from_PDEM.py in the Data_preparation folder can be used as references.
Then, you can follow the scripts in the Train_and_Test folder to train and test the model using your own data.