Peak Clipping Research Articles

Intelligibility of Individual Consonant Sounds Distorted by Infinite Peak Clipping

Previous experiments on infinite peak clipping, using whole-word intelligibility as criterion, have shown that speech may retain high intelligibility when relative amplitude information in the waveform is eliminated. To determine whether this effect holds true for all speech sounds or whether individual phonemes are affected to a different degree, in the present experiment 10 listeners phonetically transcribed infinite-peak-clipped nonsense words tape recorded by three talkers. The test materials were based upon a corpus of six plosive and six fricative consonants, voiced and voiceless, combined with five vowels in VCV utterances. Eight syllable types containing the twelve consonants provided a controlled context with equal response alternatives for each stimulus. A differential effect of clipping was evident: dental fricatives were severely degraded in intelligibility, while plosives and alveolar fricatives tended to remain relatively unaffected. Intelligibility of voiced versus voiceless sounds was not differentially affected by clipping. Confusion matrices revealed that most errors occurred within the same voicing or manner-of-articulation category.

Effects of Differentiation, Integration, and Infinite Peak Clipping upon the Intelligibility of Speech

Previous experiments on the effects of distortion in voice communication circuits have shown that intelligibility is impaired surprisingly little by the type of amplitude distortion known as peak clipping. It has been found, in fact, that conversation is possible even over a system that introduces “infinite” peak clipping, i.e., that reduces speech to a succession of rectangular waves in which the discontinuities correspond to the crossings of the time axis in the original speech signal. The intelligibility of the rectangular speech waves depends critically upon the frequency-response characteristics of the speech transmission circuits used in conjunction with the “infinite clipper.” In the present experiments, resistance-capacitance circuits with sloping frequency-response characteristics (tilting circuits) were introduced into the system at points preceding and/or following the clipping circuit. The interactions of the nonuniform frequency characteristics of the resistance-capacitance circuits with the nonlinear amplitude characteristic of the clipping circuit were studied by means of articulation tests. That there was strong interaction is evidenced by the fact that word articulation scores of 97 and 15 percent were obtained with two systems consisting of the same components in different orders. The components were (1) a tilter with a frequency-response characteristic rising 6 db per octave (a “differentiating” circuit), (2) an infinite peak clipper, and (3) a tilter with a frequency-response characteristic falling 6 db per octave (an “integrating” circuit). When these distorters were cascaded in the sequence 1-2-3 the speech output consisted of triangularly shaped waves which sounded very much like normal speech and which were highly intelligible (97 percent). When the reverse sequence (3-2-1) was used, the speech output consisted of sharp pulses giving rise to extremely poor quality and very low intelligibility (15 percent). Tests with single distorters and with pairs of distorters indicated that: (1) In the absence of frequency distortion, infinitely clipped speech is of poor quality but moderate intelligibility (50 to 90 percent depending on the listeners skill and familiarity with the test words). (2) A differentiator or an integrator preceding the clipper determines the degree to which intelligibility is impaired by infinite clipping. (3) A differentiator or an integrator following the clipper (or used alone in a linear system) affects the quality but not the intelligibility of the speech transmitted by the system.