Social Robotics in Therapy of Apraxia of Speech

Table 8

Numbers summarizing the experimental conditions and results of our proposal.

Recognized poses	3 poses (“a,” “u,” mouth closed)
Dataset instances	1625 for experiment 1 and 13497 for experiment 2
Input features per instance	54 features: 18 mouth points 3 components (x, y, z)
Users involved in the dataset	14 users for the 2 datasets
Classifiers tested	4 classifiers
Metric for comparison	Macroaverage F-score
Best algorithm for classifying poses	C4.5 (0.81 and 0.95 of macroaverage F-score in the experiments)
Users involved in the real tests	7 users