MonitVox : prova de conceito de uma solução mHealth rastreável para triagem, monitoramento e análise vocal com processamento acústico, computação em nuvem e inteligência artificial

Abstract

Voice assessment involves auditory-perceptual, self-reported, and acoustic dimensions whose isolated use may hinder result integration and process traceability. This thesis aimed to develop and technically evaluate an integrated mHealth solution, named MonitVox, to support vocal screening, monitoring, and analysis by combining self-assessment, standardized audio recording, VAD-based acoustic processing, auditory-perceptual judgment, vocal indices, operational recommendation, and computational models. MonitVox was implemented as a distributed architecture composed of an Android application, a Spring Boot backend API, and a Python analytical component, supported by relational storage, authentication, and messaging services. In the primary operational workflow, the system incorporates the Brazilian Dysphonia Screening Tool (Br-DST) as the self-reported component, the Brazilian General Grade of Vocal Deviation Index (IGGBr) as the acoustic component, and the MonitVox matrix as an interpretable recommendation mechanism for patients. The matrix was applied using IGGBr obtained from VAD-based acoustic processing and showed occupation of all four operational outputs and a positive ordinal association with the observed general grade in the retrospective analysis. In the specialist-dependent technical workflow, the system runs the Integrated Vocal Deviation Index (IVDI) and a combined index treated in this thesis as a functional verification component for the integration between IVDI and IGGBr. The results showed that voiced segmentation improved the agreement between extracted acoustic measures and the reference dataset and strengthened the acoustic axis of the matrix, although full IVDI training with VAD-processed measures did not show an unequivocal gain over the reference model in the test set. It is concluded that MonitVox is a functional and traceable mHealth artifact, with its main applied contribution centered on the operational matrix and its methodological contribution centered on the integration of mobile collection, VAD-based acoustic processing, specialist assessment, and responsible interpretation of quantitative models. The solution still depends on prospective validation, formal usability evaluation, robustness studies across devices, and additional acoustic quality assessment before broader clinical use.