This doctoral thesis investigates how Artificial Intelligence (AI), combined with speech and text analysis, can offer new tools for psychological assessment. By studying three key psychological dimensions—attachment style, emotions, and depression—it demonstrates the potential of speech as an objective marker of mental states.

Project Objective:

The main objective of this work is to develop and evaluate AI models capable of analyzing speech to detect relevant psychological patterns in three core constructs: attachment style (stable), emotions (transient), and depression (clinical). To this end, specific datasets were designed and collected, and speech processing and natural language processing methods were applied, including traditional machine learning techniques and deep learning models. Additionally, keyaspects such as model interpretability and fairness were addressed, with an emphasis on evaluating potential gender bias.

Impact:

This project makes a significant contribution to the advancement of objective and scalable tools for psychological evaluation. By offering quantifiable, speech-based methods, it can enhance the accuracy and accessibility of mental health diagnosis, supporting clinical professionals in the detection, understanding, and monitoring of psychological disorders. Furthermore, it reinforces the role of AI as a valuable ally in the development of innovative solutions for social and health-related challenges.

This project has been funded by the European Social Fund Plus (ESF+) in collaboration with the Generalitat Valenciana(GVA) through the Recruitment of Predoctoral Research Staff, grant number ACIF/2021/187.