Detecção de depressão pela fala empregando rede neurais profundas
Nenhuma Miniatura disponível
Data
2020-02-10
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Goiás
Resumo
Depression is a mental disorder that represents a major public health problem, with a 20% increase in the number of cases in the last decade. The presentation of depressive symptoms is not padronized, causing isolation and impairment in work, studies, sleep and eating. Early diagnosis remains one of the main challenges. Recent advances in machine learning methods make it possible to analyze speech, text, and facial expressions for early diagnosis and detection. This paper proposes the use of deep neural networks to detect depression, based on the patient's speech analysis, recorded during a clinical interview. For this, the pre-processing of the audios was performed, thus generating the spectrograms, mel-frequency cepstral spectrograms and the mel-frequency cepstral coefficients. These measurements were then used in the training and testing of the architectures developed here. Different combinations of network hyperparameters and spectrogram dimensions were analyzed. The results show lower root mean square error values for the application of cepstral coefficients (5.07), compared to the literature (6.50). Therefore, the potential of this method to further assist in detecting depression is envisaged. Future studies are needed to improve and validate this method applied to a sample of national data.
Descrição
Citação
MORAES, Larissa Vasconcellos de. Detecção de depressão pela fala empregando rede neurais profundas. 2020. 62 f. Dissertação (Mestrado em Ciência da Computação) - Universidade Federal de Goiás, Goiânia, 2020.