This work is about the research of stuttering which is a disease of speech. The main aim is to develop a method which is able to identify stuttered parts in the speech of a stuttering person. This can be used indirectly to improve the therapy of stuttering.
The exact definition of stuttering had to be made in order to know and classify our targets for researching them. In case of stuttering we are looking for any kind of disfluencies, which affects the natural rhythm of the speech. These are unintented repetitions, interruptions and prolongations during speech.
In order to research certain features of speech a database must be created and processed. In our case the boundaries were to be found among the voices in the speech. Besides stuttered and non-stuttered parts were also separated in the samples. The stuttered parts were given categories according to the classification of stuttering.
Using these values a research could be conducted of the identification of suttering. Five algorithms were implemented and tested. The first algorithm tested the accuracy of an automatical method which was used to segmentate the phonemes in the samples. The other algorithms extracted different features from the vowels of speech. These features were tested whether their values were different in normal and stuttered speech.
The aforementioned algorithms were implemented in MATLAB in the form of a library. After executing them their outputs were to be compared using statistics. First the outputs of every speaker were compared separately. Then the data was summarized among the speakers in order to conduct the comparisons with more data between the subsets.
Finally some methods were described about the possibility of finding stuttering automatically. Therefore suggestions were also given how to proceed with the current work on.