EGRA-Xhosa-14.9k: Annotated Child Reading Audio Dataset

The project involves collecting the child reading dataset for the language is Xhosa, a South African Bantu language. The collected dataset is then processed with the help of native speakers and utilized to train state-of-the-art machine learning models focussed on assessing whether the child has spoken the word correctly or not.
The dataset contains 14,972 recordings with an average of 4 seconds each. Each recording is annotated by three independent markers and consists of children speaking a particular word or letter from the Xhosa language in a classroom setting.
Please note that the attached zip file contains ~14,000 files. If you download this file to a Onedrive or Sharepoint location, you may be affected by the 10,000 files limit to download. When unzipping or downloading, take care to ensure that all the files are downloaded completely.