Computer assisted assessment of
children with Speech Disorder
Shanghai JiaoTong University & Children's hospital of Shanghai
2023.10-2024.8
Journal Paper （under review）contact me for more information
Method
Children dysarthria
automatic assessment of dysarthria
children’s dysarthria speech dataset
computer-assisted medicine
Medical industrial intersection design
Introduction
Dysarthria is a common speech disorder that hinders children's communication and development. Due to limited access to professional medical resources, many children miss the ideal time for speech therapy. Although interest in automatic speech rehabilitation has grown, there is still a lack of systems that accurately identify mispronunciation causes and provide personalized therapy. This study aims to address this issue, focusing on Putonghua (Standard Mandarin) rehabilitation, particularly syllable substitution.
Method
We analyzed clinical data from 103 children with dysarthria, identifying common pronunciation errors and appropriate rehabilitation strategies. A dataset of 72 children was created to train machine-learning models using FBank, spectrogram, and MFCC features with CNN10, TDNN, and Res2Net architectures. In addition, we compared the automated assessments with manual evaluations from professional speech therapists to validate the system's accuracy.
Results
The tests showed that the combination of FBank features and the CNN10 network achieved the best performance, with an accuracy of 96.892%, precision of 97.052%, recall of 96.894%, and an F1 score of 96.917%. These results highlight the system's effectiveness in diagnosing and categorizing speech errors, particularly in Putonghua syllable substitution.
Conclusion
This study presents an automated system that aids speech therapists in assessing dysarthria. It offers timely, personalized therapy advice and supports high-quality remote rehabilitation, addressing the shortage of therapists and high treatment costs. This system could significantly improve access to speech therapy for children, enabling effective at-home rehabilitation.
Prompt a treatment-oriented classification method for automatic assessment
DSR system
Confusion Matrix
