Feature engineering and machine learning for computer-assisted screening of children with speech disorders

Area of Science:

• Speech-language pathology
• Computational linguistics
• Machine learning

Background:

• Auditory perceptual analysis (APA) is standard for diagnosing childhood speech disorders but suffers from variability and transcription limitations.
• Automated methods are needed to objectively quantify speech patterns for improved diagnostic accuracy.

Purpose of the Study:

• To investigate the use of Landmark (LM) analysis for automatic detection of speech disorders in children.
• To introduce and evaluate novel knowledge-based features alongside existing LM features for enhanced classification.

Main Methods:

• Utilized Landmark (LM) analysis to extract acoustic events related to articulatory movements.
• Developed and incorporated novel knowledge-based features.
• Compared linear and nonlinear machine learning classifiers on raw and proposed features.

Main Results:

• The study systematically compared various machine learning techniques.
• Evaluated the effectiveness of novel features in distinguishing children with speech disorders from typical speakers.

Conclusions:

• Landmark (LM) analysis, enhanced with novel features, shows promise for automatic speech disorder detection in children.
• This approach offers a potential solution to the variability and limitations of traditional diagnostic methods.