An approach to automatically grading syllable pronunciation quality for speech rehabilitation by combining several binary classifiers into an ensemble. Time-series distance measures between a patient's utterance and a reference (such as dynamic time warping, longest common subsequence, and edit distances) serve as features, and expert quality ratings serve as labels. Students learn how heterogeneous base classifiers can be blended with a mixture method to modestly improve accuracy over any single classifier on a small medical dataset.
The article describes an attempt to apply an ensemble of binary classifiers to solve the problem of speech assessment in medicine. A dataset was compiled based on quantitative and expert assessments …