Toward the development of software to classify speech sound disorders

Abstract

The detection of children with speech sound disorders is a research area widely explored in many languages. Early detection and treatment positively affect the child's learning development.   This work aims to present a part of a project called ROMULO, which allows the evaluation, diagnosis, and treatment of children with speech sound disorders in Catalan where standard techniques based on signal processing cannot be applied due to the lack of public corpus. Specifically, we present an error classifier developed with a rule-based method, to detect speech problems codified in text at the word, syllable, and segmental levels. The approach can efficiently detect and classify speech sound errors, helping the speech therapist automate this process. Preliminary results show promising results on the detection, reaching an average success rate of 74.6%.