Analiza zvočnih posnetkov glasnega branja za presejalni test disleksije

Disleksija spada med specifične učne težave in je genetsko pogojena. Zgodnje prepoznavanje prvih znakov je zelo pomembno in v diplomskem delu smo pokazali, da lahko z različnim izborom atributov in algoritmi to tudi storimo. Različni računalniški sistemi bi lahko že danes hitro in objektivno odkrili rizične znake za nastanek disleksije v zgodnjih letih in tako bi lahko otrokom nudili ustrezno pomoč že v začetku šolanja, ko težave še niso tako izrazite. V tem diplomskem delu se osredotočimo na prepoznavanje oseb z disleksijo z analizo podatkov pridobljenih iz transkripcije zvočnih posnetkov glasnega branja. Med seboj primerjamo in analiziramo različne algoritme in metode strojnega učenja in podamo rezultate. V nalogi smo ugotovili, da z različnimi algoritmi strojnega učenja že na manjšem številu primerov v učni množici dobro napovemo nagnjenost k disleksiji. Dyslexia is a specific learning difficulty that is genetic in origin. It is very important to predict the predisposition to dyslexia at an early age. In this thesis, we show that we can predict dyslexia with a different attributes and machine learning algorithms. Various computer systems could quickly and objectively detect the predisposition to dyslexia at an early age and thus be able to offer children appropriate help at the very beginning of their education when the problems are not yet noticeable. In this thesis, we focus on the identification of children with dyslexia by analyzing data obtained from the transcription of audio recordings of reading aloud. We compare and analyze different algorithms and machine learning methods and give results. In the thesis, we found that we can predict the predisposition to dyslexia well with different machine learning algorithms even with a small number of cases.