Human speech is one of the most effective means of communication. That is why the research of acoustic parameters of speech sounds is really important and can help in the recognition of phonation diseases, thus it may mean a serious breakthrough in the field of medical diagnostics.
In my thesis I am examining the automatic separation of different groups of diseases. I started my work with studying the biography of the topic. Later on I decided which groups of diseases to examine and I chose functional dysphonia and vocal cord paralysis. I selected those voice samples with which I worked, including altogether 53 patients with functional dysphonia and 49 with vocal cord paralysis.
My next task was producing the input vectors of SVM (Support Vector Machine). First, I segmented the samples then I determined their acoustic parameters. For coaching and testing I used the standard deviation and the average of jitter ddp, shimmer dda, HNR and MFCC acoustic parameters.
The next step after producing support vectors was the coaching of SVM. First I examined the ’e’ sound of continuous speech. I checked the accuracy of the classifier created this way by cross validation and normal testing. After this, I did the coaching procedure of SVM with acoustic parameters related to the whole speech and I used full cross validation for testing.
My best separation result during the semester was 81.37%, which can be considered to be an outstanding result according to my research. This accuracy was achieved while examining whole speech using standard deviation and averages of jitter ddp, shimmer dda and mfcc1 acoustic parameters.