Pitch continuity and speech source attribution.

Two experiments showed that continuity of the pitch contour is an important perceptual indicator that a formant pattern comes from a single speech source. Subjects listened to repeatedly played formant patterns that changed smoothly between two vowels. When the pitch was a monotone, these patterns were heard as containing semivowels and liquid consonants; but when a discon-tinuous, steplike pitch contour was imposed on the patterns, they divided into two perceptually distinct speech sources, and the phonemic percept changed to a predominance of stop consonants. In the second experiment, this effect was shown to be due to changes in pitch itself, rather than to concomitant changes in amplitude or energy. Earlier suggestions that continuity of both formant structure and pitch contour are important for the perceptual co-herence of speech were borne out. When one listens to someone speaking against an irrelevant background of other voices, how does one know which sounds are part of the attended speech and which form the background? Differences in spatial location (Sayers Cherry, 1957) and ampli-tude (Egan, Carterette, Thwing, 1954) help the human listener to attend selectively to one voice among many, and a common pitch allows one to bundle together the individual formants of a single speaker (Broadbent Ladefoged, 1957; Parsons, 1976). Although these dimensions of the speech wave are useful for sorting the elements of sound present at any one in-stant into potential gestalten, the listener must make additional assumptions to de-termine which bundles from successive moments of time belong to the same sound source. Speakers change their location, loudness, and voice pitch from moment to moment, so some rule must be used by a listener for these dimensions to be used successfully in natural situations. This problem of the temporal coherence