Wydawnictwo Medyczne Mediton

Under the scientific patronage of

the Polish Society of Audiology and Phoniatrics


Original articles

« return
Mel cepstral analysis of voice in patients with vocal nodules
Ewa Niebudek-Bogusz 1/, Jacek Grygiel 2/, Paweł Strumiłło 2/, Mariola Śliwińska-Kowalska 1/
  • 1/ Klinika Audiologii i Foniatrii, Instytut Medycyny Pracy w Łodzi
  • 2/ Instytut Elektroniki, Politechnika Łódzka w Łodzi

It has been shown that acoustic analysis could be a valuable objective tool for diagnosing of dysphonia. Recently, special attention is paid to mel cepstral domain, which is obtained by transforming nonlinearly the frequency scale of the analysed voice signals.

To assess the applicability of Mel Frequency Cepstral Coefficients (MFCC) in acoustic analysis of voice in patients with vocal nodules.

Materials and methods.
The study comprised 40 female controls with normal voice (mean age 43 years) and 40 females with vocal nodules (mean age 41 years). The acoustic testing involved recording sustained vowels ‘a’ and four standarised sentences. The voice samples were analyzed by computed analysis of MFCCs in which the Sammon projection was employed for data dimensionality reduction and the Support Vector Machines for data classification.

The comparison of MFCC results revealed significant differences between study group with dysphonia due vocal nodules and controls. The following diagnostic test performances for patients with vocal nodules were obtained: 1) 72% sensitivity and 70% specificity for the sustained vowel, 2) 87% sensitivity and 90% specificity for the standardized sentences (averaged for 4 sentences). The results presented better accuracies for the standardized sentences. This is in agreement with a recent finding that continuous speech provides richer information about harmonic disorders in comparison to sustained utterances of vowels.

The mel cepstral-based analysis of the voice samples can be a promising tool for estimation of harmonic disturbances in pathological voices. Further studies on larger groups of subjects are required.

Otorynolaryngologia vol 10. no 4. Decemberpages: from 176to 181
full version