Characterization of Dysphonic Voices by Means of a Filterbank-Based Spectral Analysis: Sustained Vowels and Running Speech

摘要: Summary Objectives This article presents a comparative study of the spectral power distribution for normal and dysphonic voices, both sustained vowels running speech. The objective this was to find robust cues dysphonia in domain. For purpose, recordings from two databases are processed, one them including Additionally, new measure stability is introduced (decorrelation time). application spectrum also tested as cue dysphonia. Materials Methods analysis done having an auditory model filterbank approach references computation discrete spectrograms. Results obtained three sets belonging different databases. reported results indicate that only minor differences exist shape voices when performing vowel phonation tasks. However, calculated band decorrelation times bands between 2000 6400Hz significantly less stable voices. As speech, not such good indicator dysphonia, but there significant difference level high-frequency (above 5300Hz). In addition, means sampling rates above 10.6ksps needed assessing speech Also, involving short-time analysis, frame 100 frames/s should be preferred.