Facial Feature Points for Japanese Speech Content Estimation

An automatic minute generating system based on speech recognition technology can be used to improve the efficiency of minute meetings. When only one microphone was used for speech recognition, the number of devices used was reduced, and the preparation cost and convenience improved. However, the speech recognition accuracy decreases because of the influence of noise as a disadvantage. By contrast, lip movements have characteristics that are unique to the uttered contents and can be acquired even in noisy environments. The utterance content estimated from the lip movement can correct the voice recognition content. In this investigation, features for estimating utterances were analyzed with the aim of improving the speech recognition accuracy of an automatic minute creation system. The features for estimating Japanese syllables were selected and evaluated based on the movements of the feature points in the lips and jaw.