Vinícius J. D. Vieira, Rafael R. Pertum, Renato Candido. Otimização de medidas MFCC em cenas acústicas para classificação de patologias laríngeas. Simpósio Brasileiro de Telecomunicações – SBrT’2025, 2025, Natal, Anais do SBrT 2025, 2025, pp. 1–5. (in portuguese)

Abstract

In this work, a fine-tuning of the acoustic measure MFCC (Mel-Frequency Cepstral Coefficients) is performed in different scenarios in the context of laryngeal pathologies classification. The acoustic scenes considered are: environments with and without the presence of reverb and noise (with isolated and mixed effects), and the individualized classification by gender. The pathologies considered are: Reinke’s edema, carcinoma, leukoplakia, laryngitis, polyps and vocal fold paralysis. The classifier used is based on quadratic discriminant analysis. The results indicate that there is an optimal configuration of this measure, which provides the highest accuracy values in the experiments. Furthermore, it is observed that the use of a dedicated classifier by gender provides a relevant gain in accuracy in relation to the result obtained with the generalist classifier.

Keywords

Keywords Speech signal processing, laryngeal pathologies classification, MFCC, optimization.

Downloads

• PDF