In recent decades, in addition to the pathological point of medical perspective, a preventive approach was formed, i.e. to prevent their formation rather than the treatment of disease. Besides prevention, early diagnosis also got in the center of attention.
Prevention is an important thing also in the ear, nose and throat (ENT) department. One of the main manifestations is the anti-smoking campaigns, which is the cause of many respiratory diseases.
In the communication ruled world of our time, the health of the respiratory system is important, for the normal phonation. In case any difficulties occur in phonation, so in our communication, it affects our private life, and our professional position, livelihood also.
During my work, I continued to work at the BME, TMIT, Laboratory of Speech Acoustic group, in their project of the acoustic diagnostic of pathological lesion.
The aim of the project is to create an automatic, hoarse determinative system, which uses continuous speech. For achieving, the objective of this semester was to improve the previous semester’s models. I analyzed the expert’s ratings consequency. I searched for new acoustic parameters in the literature.
These parameters define the signal to noise ratio and breathlines of the signal like the variations of GNE (Glottal to Noise Excitation ratio), GQ (Glottis Quotient), IMF (Intrinsic Mode Functions) and the nonlinearity like RPDE (Recurrence Period Density Entropy) and DFA (Detrended Fluctuation Analysis). I extracted the new parameters from an extended database, examined their correlation, and usability from the model building point of view. I got to the conclusion that some parameters doesn’t behave like expected. However a narrow scope of the parameters might be useful for measuring hoarseness, in the continuous speech application also. The feature selections on the extended parameter list are more promising than the ones that only use jitter shimmer, hnr and mfcc1 in the model. Based on the early results of the cluster analysis, it might be a useful to separately analyze the women and men.