Automatic estimation of the number of minimal language units by articulation

The presented work is dedicated to the automatic analysis of the paraverbal component of human communication. The article describes systems that determine the number of minimal linguistic units (syllables and phonemes) in spoken language based on video data. Such systems can be used to assess the subject speech rate, which can be applied in the preclinical diagnosis of certain pathological conditions or determining emotional status. To conduct the research, an existing database of English words was modified, and annotations containing information on the number of syllables and phonemes in each word were obtained. During the study, a word recognition system was adapted to solve the stated task, and a new neural network architecture to determine the number of syllables and phonemes in a word was designed. The effectiveness of the developed systems was assessed on both sets of previously known to the systems words and on new words. As a result of the research, a system that determines the number of minimal language units in a spoken word was obtained, providing the opportunity for subsequent assessment of the subject articulation rate.