Advanced Time-Frequency Analysis and Machine Learning for Pathological Voice Detection
Voula Chris Georgopoulos · 2020
This paper discusses an advanced positive time-frequency analysis method with accurate zero-, first- and second-order moments combined with machine learning for detection of pathological voice signals. The time-frequency analysis is based on the Wigner-Ville distribution. The transfer learning approach is used to train a convolutional neural network. Specifically, the GoogLeNet network is trained using the normal and pathological voices of the KAY database and the results are presented.