Vocalyze: A Deep Learning Approach to Detecting Major Depressive Disorder (MDD) and Tracking Treatment Efficacy via Vocal Acoustic Inflections and Sentiment Analysis

Overview

Major depressive disorder (MDD) stands as one of the most prevalent and perilous mental health conditions globally. Nearly sixty percent of individuals who have lost their lives to suicide display symptoms of MDD. However, current diagnostic measures rely heavily on doctor evaluations or ambiguous, non -standardized surveys, lacking biomarker analysis. Furthermore, treatment evaluations occur sporadically, with little effective monitoring in most cases. However, MDD presence and severity can be identified an d tracked by analyzing vocal prosodic features, and contextual semantic analysis of spoken words. Efficient diagnosis and monitoring of MDD is achievable through Vocalyze, a user -friendly portal. Patients provide a vocal response to a provided prompt to diagnose MDD and later provide daily responses in the Vocalyze UI to assess treatment efficacy. The system is built on two models. Model 1 is a convolutional neural network that provides a diagnosis and predicts PHQ -9 scores based on vocal prosodic features. This model was trained with 187, 276 data sets, tested with 46,819 data sets, and produces a diagnosis accuracy of 98% and a r -squared score of 0.921 between predicted and ground truth PHQ-9 scores. Model 2 is a natural language processing model and performs sentiment analysis to analyze transcripts of the daily recordings, consistently producing a confidence level for each sentiment over 5%. Vocalyze, through effective analysis of vocal prosodic biomarkers possesses an accuracy rate surpassing current met hods by over 40%. This system is an accurate, accessible, and cost-effective alternative, countering current dangerous practices contributing to alarmingly elevated suicide rates.