Phonological Feature-based Automatic Pronunciation Analysis and Feedback System

Case ID:
Web Published:


Speech sounds disorders are common in foreign language learners and school-age children and current effective treatments rely largely on prolonged in-person speech therapy that is unavailable to many due to location or resources. Oregon Health & Science University (OHSU) researchers have developed a computer-assisted pronunciation training program that could reduce the cost and expand the availability of pronunciation training.


Technology Overview

Speech sound disorders affect roughly 10% of school age children and these disorders can adversely affect communication, academic performance and interaction level. The most effective treatments for speech sound disorders relies on prolonged practice and training with a speech language pathologist, but many children and adults lack the resources and availability for this therapy.  Computer-assisted pronunciation analysis and trainings tend to have low accuracies, especially in children who have increased variability in speech patterns.  OHSU researcher Dr. Kain and colleagues have developed a program that utilizes both phoneme and phonological features to improve accuracy in detecting mispronunciations. Testing of the program in 90 children (aged 4-7) found that the program identified differences between typically developing and speech disordered groups that corresponded with differences observed by human experts.  Testing further revealed that the program could detect mispronounced phonemes with 97% accuracy in adults and 77-80% accuracy in children.  Computer-based identification of mispronunciations can allow for at-home training and specific feedback to improve patient speech.  This system may also allow for automation of clinical screening tests to increase access and diagnosis of these disorders. Utilization of this program has the potential to provide children and adults with effective tutoring on demand, at low cost, and independent of location.



A. Kain, A. Roten and R. Gale, "Diacritic-Level Pronunciation Analysis Using Phonological Features," ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 8084-8088


Licensing Opportunity

Available for licensing.



Patent Information:
For Information, Contact:
Arvin Paranjpe
Senior Technology Development Manager
Oregon Health & Science University
(503) 494-8200
Alexander Kain
Amie Roten
Diagnostics - Other
Education & Training - Speech & Language
Software - Diagnostic/Detection
© 2023. All Rights Reserved. Powered by Inteum