Assessing and Treating Speech Motor Impairment:

A Machine Learning Approach

 

Jun Wang, PhD

Callier Center for Communicative Disorders

University of Texas – Dallas

 

Thursday, February 28, 2013 at 11:00 in Room C141 Speech & Hearing

Although there is a critical need for objectively assessing the speech motor decline due to neurological diseases (e.g., amyotrophic lateral sclerosis, ALS) and for assistive technologies for people with speech impairment, few options are available due to the logistical difficulty of tongue movement recording. This research investigated novel approaches for assessment and treatment for speech motor impairment, which leverage the recent advances of 3D motion capture technologies and the power of machine learning algorithms for data analysis. Machine learning classifiers (e.g., support vector machine and Procrustes analysis) were used to classify and quantify the articulatory distinctiveness of phonemes, words, and sentences based on tongue and lip movement time-series data. A novel measure, articulatory vowel space area, was derived and used for assessing the impaired speech motor control due to ALS. In addition, an online word and sentence recognition algorithm from continuous articulatory movement data was evaluated. Experimental results showed the potential of the machine learning algorithms for analyzing tongue movements for speech assessment and for the development of a silent speech recognition system that generates speech in response to silently produced articulatory movements.