Journals / Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi / 2012 / Cilt: 14 - Sayı: 40
COMPUTER BASED VOICE ANALYSIS ON MEDICAL DIAGNOSIS
- Pages
- 11–22
- DOI
- —
Abstract
The change in voice quality is affected by many of voice disorders that coming from pathogical conditions of voice generation organs. The aim of this study is to help that the clinicians could be diagnosed about voice disorders with non-invasive based analysis. In our work, amplitude perturbation quotient, pitch period perturbation quotient, degree of unvoiceness, Teager Energy Operators averages of wavelet transform coefficients, and higher-order statistics parameters have formed the feature vectors. The voice segments belonging to different pathological or normal classes were classified by backpropagation based multilayer perceptron networks. In backpropagation based multilayer perceptron networks, resilient, scaled-conjugate gradient, and Brodyen-Fletcher-GoldfarbShanno learning algorithms were used in training. According to the results of the simulation studies, scaled-conjugate gradient algorithm gave the best results.
Özet
conditions of voice generation organs. The aim of this study is to help that the clinicians could be diagnosed about voice disorders with non-invasive based analysis. In our work, amplitude perturbation quotient, pitch period perturbation quotient, degree of unvoiceness, Teager Energy Operators averages of wavelet transform coefficients, and higher-order statistics parameters have formed the feature vectors. The voice segments belonging to different pathological or normal classes were classified by backpropagation based multilayer perceptron networks. In backpropagation based multilayer perceptron networks, resilient, scaled-conjugate gradient, and Brodyen-Fletcher-Goldfarb- Shanno learning algorithms were used in training. According to the results of the simulation studies, scaled-conjugate gradient algorithm gave the best results