Electronic Binary Selection System for Phoneme Classification
Jacob Wiren, H. L. Stubbs · The Journal of the Acoustical Society of America · 1956
A successive-binary-selection system for automatic classification of spoken English into several groups of phonemes is described. The first step separates voiced from unvoiced by measuring the filtered output in the range of the pitch frequency. The next step separates turbulent (noise-like) from nonturbulent by measuring the filtered output in the range of the first formant. Nonturbulent sounds are classified into 6 groups of 2 or 3 phonemes each in successive selections based on spectrum measurements. For each of the two groups of phonemes being separated in a given step, a frequency band is chosen where phonemes of that group have large amounts of energy compared to those of the other group. The polarity of the weighted difference of the corresponding band-pass-filter outputs determines the classification. Unvoiced turbulent sounds are separated into stops and fricatives by means of the amplitude information contained the onsets of these sounds. Two of the fricatives, |s| (see) and |∫| (she), are separated by means of their zero-crossing-density levels.