An Algorithm for Segment Durations in a Reading Machine Context.

摘要

The purpose of this study was to devise an algorithm for the acceptable control of segment durations in a reading machine. The general experimental method was as follows: Subjects were asked to read lists of words and phrases from 3 x 5cards. The results were recorded,and narrow-band and broadband spectrograms were produced for each utterance with the use of a Kay sonograph. By using segmentation rules,the duration of each segment in each utterance was measured and tabulated. From these data, a model was devised which sought to predict the experimental results for each subject individually. Several interesting results were obtained from these studies. First,it was discovered that the duration of a vowel could be modeled as a function of the word-level stress of the syllable in which the vowel appeared,the structural location of the word in which the vowel appeared,the number of syllables in the word,the consonant context of the vowel,the vowel's location in the phrase,and the proximity of any word or syllable juncture. Second,it was shown that,in general,consonants outside of clusters do not vary much in duration. Third,consonant durations could be shown to vary greatly in clusters. In general,no results were available for the behavior of liquids or glides. (Author)