Classification of reading levels of primary school students by artificial intelligence on sound

By supporting the customizable education and training process, artificial intelligence can tailor educational environments to individuals' needs, capacities, interests, and abilities. Recent research shows that the strategic value of artificial intelligence for education is increasing more and more. Artificial intelligence can be a learning tool that reduces the burden of both teachers and students and provides effective learning experiences for students. For this purpose, in this study, 20 primary school students were asked to read a fixed reading text and take a voice recording. These audio recordings were labeled by the teacher as good, medium and bad reading levels. All sounds are first fragmented so that each sentence is a separate sound file. Thus, a total of 449 sound files were obtained. Then, a 256-column feature vector was created from these audio files with the Local Binary Pattern method. Various classification algorithms are used on this feature file. The highest success rate was obtained from the Cubic SVM algorithm with a rate of 77.5%. With this study, it is thought that students can observe their own reading levels and increase their level by reading more. It is possible to expand this study by using a larger dataset other than the dataset here. In addition, the success rate can be increased by using different feature extraction methods.