Project DetailsThis work focused on creating a platform for the Koren Center, which supports children, teens, and adults with reading and writing difficulties. The Koren Center’s methodology involves students reading aloud short text segments displayed on a screen, while the teacher listens and marks the errors for himself. Our goal was to develop a system capable of detecting reading and pronunciation errors autonomously, without requiring a human supervisor. The main errors defined by the Koren Center are: "omission of words", "change of word order" and "pronunciation errors".
As part of the project, As part of the work, we received examples of recorded text segments, some of which are correct and some of which contain errors. These recordings were processed using Whisper software, which transformed them into text. This text was compared to the original text of the story using code we wrote. For each text segment, the code identifies whether it contains one or more errors, and classifies the errors according to the desired categories.
Throughout the project, we overcame various challenges in error detection and in achieving accurate speech-to-text conversions. After significant effort, we achieved an accuracy rate in which roughly 90% of sentences with errors were detected, with a comparable rate in error classification. This high degree of precision enables Koren Center to rely on the platform’s results, ultimately easing the teacher’s workload by reducing the need for direct supervision.
