SBIR Phase I: Adaptive Phonetic Analysis for Personalized Dyslexia Reading Support

The broader commercial impact of this SBIR Phase I project lies in addressing dyslexia, a condition affecting millions that creates significant educational, economic, and social barriers. Learners with dyslexia often struggle to develop foundational reading skills, leading to disparities in academic achievement and limited opportunities. This project addresses these challenges by developing an innovative, technology-driven solution that enhances decoding ability through customized reading practice.

By integrating advanced speech recognition technology with evidence-based teaching methods, the solution provides personalized, scalable, and affordable learning experiences tailored to the needs of dyslexic learners. The technology identifies reading errors with unprecedented granularity, generating adaptive learning content to empower learners to overcome obstacles and build fluency.

This initiative has the potential to improve decoding skills for hundreds of thousands of at-risk readers, helping to close educational gaps and create equitable opportunities. Beyond individual benefits, the societal impact includes fostering academic confidence, increasing graduation rates, and enhancing employability, contributing to the well-being of communities.

Commercially, this project positions itself as an innovative educational solution, offering tools to help learners and institutions address one of the most pervasive barriers to literacy and lifelong success.

This Small Business Innovation Research (SBIR) Phase I project brings reading error identification to an unprecedentedly fine granularity and generates engaging practice contents on an adaptive difficulty level to improve dyslexic readers’ decoding ability. Dyslexia often causes reading errors including insertions, deletions, substitutions in phonemes, which are mostly autocorrected by traditional word-level automatic speech recognition (ASR) technology.

Additionally, personalized reading contents for dyslexic readers is expensive to generate on a scale. Therefore, to achieve ideal error tracking accuracy and provide affordable, accessible and personalized reading contents to dyslexic population, this project proposes to: (a) build a grapheme-phoneme extension of existing phoneme-only pronunciation dictionary, (b) train an accurate phonetic ASR specifically devoted to transcribing dyslexic readers’ utterances that preserves reading errors, (c) refine the downstream error tracking solutions to identify decoding challenges down to grapheme-phoneme level, and (4) complete a customized story generation pipeline to create easily decodable and interest-infused reading contents.

At its accomplishment, this project will achieve above 97% coverage of the original pronunciation dictionary, below 15% phonetic error rate for its phonetic ASR component, below 10% false alarm rate for its error tracking model, and above 95% success rate for its customized story generation pipeline.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.