The Annual Status of Education Report (ASER), a nationwide survey of reading and math achievement of children from rural India, has been conducted annually since 2005. The ASER test inference is about a child’s level of foundational reading skills (letter identification, word decoding, proficiency in reading paragraph and story) and basic math ability (number recognition, subtraction, and division). The tests are orally and individually administered and require about 10 minutes of administration time. The tests are designed to understand what students can do and the skills they have mastered. More info at: https://www.asercentre.org/
The ASER dataset was collected with the primary objective to automate the ASER process. This will help save the cost and time involved every year in training the volunteers and conducting the test. The dataset has been recently published in ICASSP 2020:
D. Agarwal, J. Gupchup and N. Baghel, “A Dataset for Measuring Reading Levels In India At Scale,” ICASSP 2020 – 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 2020, pp. 9210-9214, doi: 10.1109/ICASSP40776.2020.9053380.
Recent research in deep learning and speech recognition technology has made it possible to develop computer-based reading evaluators that listen to students, assess their performance, and provide a score or immediate customized feedback.
In this challenge, you are provided with a subset of the ASER Dataset. You are provided with multiple audio files of children trying to identify and reading out different English letters (small and capital letters). The task is to classify whether the child could recognize the letter correctly or not after analyzing the audio file for the same.
The goal in this competition is to take audio of a child recognizing letters of English (small/capital), and determine if the child could recognize it correctly or not.
For every audio file in the test set, you should predict the correct label.
This competition is evaluated on the categorization accuracy and precision of your predictions (the percentage of audios you get correct).
More details about the dataset, including the baseline solution that has been tried on it and their levels of success, can be found at PrathamOrg/ASER-Dataset
The dataset is made available under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.