Omar A.2024-12-172024-12-172022https://repository.sdu.edu.kz/handle/123456789/1580We are interested in developing software that can automatically recognize particular phone segments that a non-native student of a foreign language has pronounced incorrectly. A language training system can provide the student with feedback about individual pronunciation errors by using the information about the phone level. For this purpose, in this work, I am trying to develop a transformer model that will recognize spell errors in speech. There were two strategies that were examined: the first one on the original audio dataset, and the second one on the synthetically augmented dataset. Both experiments were compared in this work.ensoftware, phone, segments, a non-native student, foreign language, datasetOther