| Summary: | Depression is a mental disorder that is increasingly becoming common in people’s lives and that can have serious implications on human beings. Over 264 million people worldwide suffer from this disorder, and the trend is for these numbers to increase over the years. With this in mind, it is necessary to develop depression recognition and prediction methods by analysing natural language, non-verbal behaviours and speech processing. This is an area of study with high interest, since it can support clinicians on patients diagnosis and treatments, as well as it can also serve the patients by receiving a robust diagnosis and an adequate treatment guide so they can overcome the disorder. This dissertation focuses on developing a method that can quantify and predict if a person suffers from depression, basing itself on studies and articles in the area, published in international conferences. With access to interviews and clinical data, a language and speech analysis for each participant was performed, with the intent of extracting key characteristics that could assist depression identification. After the extraction, experiments with unimodal and multimodal models were developed with the objective of quantifying depression correctly for each participant. These models outperformed the AVEC conference baseline and presented comparable results with other published models in that same conference.
|
|---|