Using speech data to accelerate progress toward early detection and prevention of reading challenges

Through ReadNet, we created an open database with unique potential for improving reading outcomes for U.S. children. This database includes tens of thousands of anonymized, annotated speech samples and direct assessment data of reading skills taken from students who were screened for risk of reading difficulties when they were in kindergarten and include the individual reading outcomes of these same students in later grades.

By sharing this powerful dataset through online modeling competitions for data scientists hosted by DrivenData, we seek to develop the best models to answer questions that will help predict future reading risk of children and improve the accuracy of speech recognition technology. We expect this will enable better tools for automated in-school screening of all children that are efficient, valid, scalable, and equitable, and also allow us to predict future reading challenges from kindergarten speech data.

Research for the ReadNet Project is being conducted through the Gabrieli Lab, Quantitative Methodology and Innovation at Florida Center for Reading Research, and Senseable Intelligence Group. It is supported by generous gifts from Schmidt Futures and Citadel founder and CEO Ken Griffin.