license: cc-by-nc-4.0
task_categories:
- automatic-speech-recognition
language:
- hi
- mr
ReadNet
Dataset Description
ReadNet is an audio dataset collected from more than 2,00,000 children in the age group of 5-16 years in Hindi and Marathi language. The dataset consists of audio files in the wav format where children read out ASER Samples which consists of letters, words, stories and paragraphs in their native language. This dataset is a subset of the larger dataset which consists of an estimated 2500 hours of data. This dataset consists of ~87 hours of audio data in Hindi and Marathi,
Languages
Hindi and Marathi
from datasets import load_dataset
dataset = load_dataset("PrathamOrgAI/ReadNet")
Data Fields
URL: URL of the audio file which can be downloaded in the .wav format
Transcribed Text: Transcription of audio files in their respective language
Annotations
Annotations for the audio files are provided in the Transcribed Text Column. The annotation of audio files was done on an annotation portal which was developed inhouse by Pratham Education Foundation
Annotation process
A team of annotators were hired for the annotation of audio samples in two languages. Training regarding the annotation portal was provided to the annotators. A dry run for the an- notation was done initially and the annotation was reviewed and feedback was given to the annotators to remove incon- sistencies and errors in annotation. Annotation guidelines are provided in the annotation portal itself, in both languages for the annotators reference. Since the scale of the data set is pretty large, we have decided to annotate some portion of the data twice and the remaining data once depending on the re- sources
Who are the annotators?
The Annotation was done by experts from the ASER team
Personal and Sensitive Information
The dataset consists of people who have donated their voice online. You agree to not attempt to determine the identity of speakers in this dataset.
Social Impact of Dataset
Acknowledgements
We would like to thank Schmidt Futures and the Sarva Man- gal Family trust for funding the ReadNet Project. We would also like to thank Dr. Wilma Wadhwa and Anil Kumar Ka- math from the ASER center for their exceptional work on planning and execution of the data collection and annotation exercise. We would also like to thank Rajarshi Singh from the PAL Network and Uday Narayan Singh who helped us with our sampling strategy. Lastly we would like to thank Dolly Agarwal and everybody from the Tech team of Pratham Edu- cation Foundation for their work in making this project pos- sible.