This paper presents a performance comparison for a voice pathology detection system dealing with different types of audio data. Several files of sustained phonation of vowel /a/, from Kay Elemetrics database, were encoded with MP3 algorithm with various bit rates (160, 48 and 24 kbps). A multilayer perceptron classifier is then used to automatically detect the normal from the pathologia files. Results are compared with those obtained for the original database, using confusion matrices and DET plots. There are no significant differences between the designed detectors.
Index Terms. Voice pathology detection, Multi-Layer Perceptrons, MPEG Audio layer 3 (MP3)