Empirical Mode Decomposition. Spectral Properties in Normal and Pathological Voices

MARÍA EUGENIA TORRES; GASTÓN SCHLOTTHAUER; HUGO LEONARDO RUFINER; MARÍA CRISTINA JACKSON MENALDI

Amberes, Bélgica

Congreso; European Medical and Biological Engineering Congress Engineering for Health; 2008

European Alliance for Medical and Biological Engineering and Science / International Federation for Medical and Biological Engineering

Empirical Mode Decomposition (EMD) is a data driven techniqueproposed by Huang. In this work, we explore spectralproperties of the empirical modes and apply them tospeech signals corresponding to real and simulated sustainedvowels. We will show that using basic pattern classificationalgorithms, the spectral properties of only three modesare enough to discriminate between normal and pathologicalvoices. We have obtained a 99% of correct classificationsbetween normal and pathologic synthetic voices; whilein the case of real voices the percentage of correct classificationswas 93:40%: Additionally we consider two pathologiesof different etiology and treatment, which, given the similarityof their voice characteristics are frequently misdiagnosedin clinical practice: muscular tension dysphonia and adductorspasmodic dysphonia. Preliminary results with a reducedreal data base suggest that this approach could provide usefulorientation to physicians and pathologists.