Relative Intelligibility of Different Transforms of Clipped Speech

Publisher Website
Google Scholar
Abstract
In a speech wave that has been subjected to infinite clipping, only the sequence of time intervals between zero crossings remains, and yet the result is fairly intelligible. If positive-going and negative- going zero crossings are signaled by pulses of either the same or opposite polarity, it is possible to distinguish the time intervals between adjacent zero crossings from those between alternate zero crossings (i.e., crossings in the same direction). It has been found that transforms that maintain some indication of the polarity of the zero crossings are the most intelligible, whereas those that make no distinction are the least intelligible. The transforms in which zero crossings of only one kind are signaled form an intermediate category. Neither the polarity of the pulses nor the polarity of the zero crossings has any great effect on the relative intelligibility. Using data obtained in an experiment measuring the intelligibility of clipped speech, a confusion matrix of the various phonemes was constructed. It was found that vowellike phonemes are less often confused than fricativelike ones.

This publication has 0 references indexed in Scilit: