Cathay General Hospital's AI screening model helps early detection of speech disorders in childrenMay 23, 2024

Speech Sound Disorder (SSD) is a common type of speech and language impairment in children, affecting their communication abilities. Traditionally, screening and assessment of SSD in children have relied on speech therapists' auditory evaluations. However, subjective assessment methods can lead to misdiagnosis or missed diagnoses. 

The team led by Director Ya-Wen Tu at Cathay General Hospital has developed the "Speech & Sound-SPEAK" app, which uses deep learning technology to assess speech disorders in children in real-time. 

Challenges and solutions in detecting speech disorders in mandarin-speaking children 

Detecting speech disorders in children is challenging due to the significant individual variations in children's speech and the developmental nature of their pronunciation. It is difficult to distinguish between normal pronunciation deviations and abnormal pronunciation errors. Additionally, the syllabic structure and phonetic characteristics of Mandarin differ from those of English and other languages, making it hard to directly apply English-based detection methods to Mandarin. 

This model reduces the need for forced alignment and phonological rules, simplifying the model design process and decreasing the reliance on domain-specific knowledge. By combining linguistic and acoustic features, it effectively addresses the challenges of detecting and diagnosing speech disorders. 

AI-powered screening: More accurate and faster detection of speech disorders 

Users of the app can use child-friendly picture cards to prompt corresponding pronunciations, and then upload the raw speech signals to a cloud server. The system employs a deep learning model to analyze the speech signals, identifying and displaying six common clinical phonological error patterns: stopping, backing, fricativization, affrication, devoicing, and final consonant deletion. This leads to more effective results in speech recognition, error detection, and phonological analysis. 

Providing diagnostic support for speech therapists 

In addition to real-time assessment of speech disorders in children, the "Speech & Sound-SPEAK" app offers valuable diagnostic support for speech therapists, including: 

  • Assessment Result Analysis: The system provides detailed analysis of the assessment results, including error rates for each phonological process, aiding speech therapists in diagnosis. 
  • Speech Education Information: The system offers comprehensive speech therapy education information to help parents understand the nature of speech disorders. 
  • Follow-Up Treatment Map: The system provides a follow-up treatment map to assist parents in finding suitable speech therapy facilities. 

Director Ya-Wen Tu emphasized that speech disorders are common in children and, if left untreated, can affect their academic performance and social interactions. The development of this app aims to help more children detect speech disorders early and receive timely treatment. 

Resource (Mandarin): 

新創焦點-國泰醫院語音AI篩檢模型 助及早發現兒童語音障礙