Towards Culturally Adaptive Mental Healthcare: A Design and Engineering Approach to Speech-Based Psychosis Detection using Deep Learning

Abstract

This paper presents a novel interdisciplinary approach to developing an intelligent speech-based system for early intervention and personalized care in mental health, specifically focusing on psychosis detection. Building upon the foundational understanding of speech as a biomarker for mental disorders, we integrate principles from design, engineering, business, and cultural studies to address the limitations of existing technical-centric solutions. Our methodology leverages advanced deep learning techniques, particularly a refined Convolutional Neural Network (CNN) model, to analyze log-Mel spectrograms derived from short speech segments, ensuring the preservation of crucial acoustic-temporal nuances. Beyond technical ecacy, this work emphasizes the critical role of user experience (UX) design in healthcare applications and incorporates cultural adaptability considerations to enhance the system's universality and acceptance across diverse populations. We detail the engineering implementation challenges and solutions, including system architecture for robust deployment and privacy-preserving mechanisms. Through experimentation, we demonstrate the system's capability to identify psychosis diagnostic status and negative symptoms, while also evaluating its performance across varied cultural contexts and assessing user satisfaction. The findings underscore the potential of a holistic, cross-disciplinary framework to bridge the gap between technological innovation and practical, human-centered mental healthcare solutions, paving the way for more accessible, effective, and culturally sensitive interventions. This research contributes to the advancement of intelligent healthcare systems by offering a comprehensive model that transcends traditional disciplinary boundaries, fostering a more inclusive and impactful application of AI in mental health.