Adjacent Context and the Intelligibility of Words Excised from Fluent Speech
J. M. Pickett, Irwin W. Pollack · The Journal of the Acoustical Society of America · 1963
Conditions of fluent utterance may so modify a spoken word that it is not intelligible when taken from context. This experiment explores certain factors in the intelligibility of such excised words. Four talkers recorded short paragraphs of English prose at 3 rates of fluent utterance: normal rate, deliberately slow rate, and deliberately fast rate. Segments containing 1–7 words were then excised from the recordings and presented to 15 listeners for identification. Intelligibility is primarily dependent upon the duration of the sample: intelligibility at a given rate improves with the number of presented words: intelligibility also improves with slower reading rates with a constant number of presented words. Samples from the fast reading rate, however, yielded somewhat lower intelligibility for equal samples in the range 200–800 msec. Relatively small effects were associated with the position of a word within the sample and its grammatical function. Implications for computer recognition of speech are briefly discussed.