Voice researchers and phoniatric doctors have been concerned for decades about how the nature of phonation disorders could be extracted from the acoustic parameters of voices.
Currently, in the health care system, people with disorders which are connected to speech producing, must wait weeks or months to get a specific diagnosis. This wasted time could increase the probability of causing an abnormality to speech producing organs (such as laryngitis), or more serious consequences such as throat cancer.
The aim of the project is to create the basics of diagnosis support system that can detect pathological speech, which uses continuous speech.
In this research, a speech database containing healthy and pathological voices, which was developed by the Laboratory of Speech Acoustic (LSA) at BUTE TMIT, has been used. During my work, acoustic parameters have been used on purpose of diagnosing speech organ's disorders. During this research, I made leave-one-out crossvalidations with KNN and SVM systems. The classifictations were made between healthy and pathological speakers for women and men separately as well.
One of the main results of my work that it worth to classify the voices into separately for women and men. From these experiments it can be concluded that the best way to classify the voice samples of healthy and pathological women and men is to use parameter selection and linear kernel SVM at the same time.
In the future we can develop a software which can help in the automatic diagnosis of different types of voice disorders if the available speech database is increased by more records of different illnesses.