Intelligibility of Excerpts from Fluent Speech: Effects of Rate of Utterance and Duration of Excerpt

Abstract

The purpose was to study factors in the intelligibility of words removed from fluent speech. The talker recorded a short text at three rates of continuous utterance: very fast, normal, and very slow. Then samples representing one or more text words were removed by gating. The same text words were removed at each rate of utterance. The samples were presented to listeners in intelligibility tests. Four talkers were tested, each with a different text from which 24 to 28 samples were removed at each rate. Short samples were less intelligible than long samples. On the average, fast and slow samples were equal in intelligibility when equated in duration. This is interpreted to indicate a balanced trade-off between slow, precise articulation of a small amount of text and rapid, slurred articulation of a large amount of text in the same time interval. Slow utterance was slightly more intelligible in noise. Finally, the samples were vocoded and then reconstructed synthetically on both a stretched time base and the normal time base. The stretch-vocoded fast samples were slightly longer in duration than the normal-vocoded slow samples representing the same words. However, the stretch-vocoded samples were lower in intelligibility.