Automatic recognition of semivowels in word context
Hiroya Fujisaki, Yasuo Satô · The Journal of the Acoustical Society of America · 1976
Based on an approximate formulation of the coarticulatory process at the acoustic level and a method for adaptation to individual differences, a scheme for reliable segmentation and recognition of connected vowels has already been established [H. Fujisaki et al., Speech Communication Seminar, Stockholm 1974]. The present paper describes a study for the extention of the scheme to recognition of vowels and semivowels in word context. Values of formant targets which represent a command for each phoneme, its duration, as well as the rate of transition between successive phonemes are extracted as parameters from time-varying patterns of formant frequencies. Among these parameters, the command duration is shown to be most effective in discriminating semivowels /j/ and /w/ from vowels /i/ and /u/. A scheme for recognition of vowels, semivowels, and their sequences based on formant targets and command duration is then proposed and tested experimentally using meaningful and nonsense words uttered by four speakers.