Identification of Unvoiced Fricative Sounds from a Clipped Speech Signal

The noise signal on the quasistationary (nontransient) sections of fricative sounds may, in its general features, be characterized by the number of zeros $2N$ on some time interval and by the scatter of the durations of the intervals between zeros. The degree of dispersion may be expressed quantitatively by the number of intervals $n_i$ and $m_i$ with durations significantly greater and significantly less than the mean duration. The reliability of discrimination of the voiceless fricative sounds on the planes of the attributes $Nn$ and $N(n+m)$ with 2240 realizations from 46 speakers (taking account of frequency of occurrence) amounted to about 91 percent. These results may be used to identify other noiselike sounds, since the properties of noiselike sounds different in the method of formation, but identical in the place of formation are very similar.