| Abstrakt: |
Parkinson's disease (PD) is a neurodegenerative condition of the nervous system that impairs movement caused by brain cell degeneration and it is most common after Alzheimer's disease seen in old people. One of the most significant symptoms in the first stages of PD is vocal impairment. As a consequence, current research has focused extensively on the possibilities for identifying PD through the analysis of speech and voice patterns. In this paper, we propose a new architecture for extracting features from audio files using mel-frequency cepstral coefficients, tonnetz representation, chromagram, spectral contrast, and mel-scale spectrograms, followed by feeding these generated features into a genetic algorithm for feature selection. Depending on acoustic features, the genetic algorithm is combined with the support vector machine to enhance the diagnosis accuracy of PD patients. The maximum classification accuracy of 98% is attained with /vowel-i/, and 90% with /grito/. The results indicate that the proposed feature combination is capable of correctly diagnosing PD in a person based only on that person's speech. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink