Facial occlusions caused by HMD (Head-Mounted Display) make capturing user's face hard. In this paper, we propose a technique to classify the mouth shapes into 6 classes using optical sensors embedded in HMD and give labels to training dataset by vowel recognition. We conducted an experiment with 5 subjects to compare recognition rates of machine learning in manual labeling and automated labeling conditions. The result shows that proposed method achieved an average of 99.9% classification accuracy in the manual labeling condition, and of 96.3% classification accuracy in the automated labeling condition.
