Key points are not available for this paper at this time.
A digit recognition system based on simple segmentation rules according to articulatory features is described. The digits are segmented according to manner of articulation, and the first two formants are automatically tracked for vowel-like segments. Each digit is represented by a sequence of segments and five formant values, namely, the first and second formant at the beginning and the end of the segment and the maximum value of the first formant within the segment. Reference patterns are established and unknown utterances are matched against them. Error rates range from 1.2% to 5.0%, and reject rates range from 1.1% to 4.4% for different experiments.
Tara Keller (Thu,) studied this question.