On the perception of intonation from sinusoidal sentences

Abstract

Listeners can perceive the phonetic value of sinusoidal imitations of speech. These tonal replicas are made by setting time-varying sinusoids equal in frequency and amplitude to the computed peaks of the first three formants of natural utterances. Like formant frequencies, the three sinusoids composing the tonal signal are not necessarily related harmonically, and therefore are unlikely to possess a common fundamental frequency. Moreover, none of the tones falls within the frequency range typical of the fundamental frequency of phonation of the natural utterances upon which sinusoidal signals are based. Naive subjects nevertheless report that intelligible tonal replicas of sentences exhibit unusual "vocal" pitch variation, or intonation. The present study attempted to determine the acoustic basis for this apparent intonation of sinusoidal signals by employing several tests of perceived similarity. Listeners judged the tone corresponding to the first formant to be more like the intonation pattern of a sinusoidal sentence than: (1) a tone corresponding to the second or third formant; (2) a tone presenting the computed missing fundamental of the three tones; or (3) a tone following a plausible fundamental frequency contour generated from the amplitude envelope of the signal. Additionslly, the tone reproducing the first formant pattern was responsible for apparent intonation, even when it occurred in conjunction with a lower tone representing the fundamental frequency pattern of the natural utterance on which the replica was modeled. The effects were not contingent on relative tone amplitude within the sentence replica. The case of sinusoidal sentence "pitch" resembles the phenomenon of dominance, that is, the general salience of waveform periodicity in the region of 400-1000 Hz for perception of the pitch of complex signals. © 1984 Psychonomic Society, Inc.