Perception of nonlinear and linear formant trajectories

Abstract

Perception of F2 trajectories in synthetic vowels was investigated. Perceptual boundaries (50% response points of identification functions) of 20-step/U-I/continua with various shapes of F2 trajectories were determined and compared with the boundary for vowels with steady-state F2. In experiment 1, the vowels were synthesized in either/j-j/or/w-w/context resulting in parabolic trajectories, then the stimuli were split into halves resulting in quadratic F2 trajectories. All stimuli were 200 ms long. For the/wVw/, and/wV/stimuli, the boundaries were at lower stimulus numbers than for the stimuli with steady-state F2, indicating that the nonlinear F2 trajectories were perceived as having frequencies beyond the extreme values actually synthesized in the stimuli. This type of signal processing has been termed "perceptual compensation." For the /jVj/ stimuli, there was only a trend for perceptual compensation of F2 trajectories. For the/Vw/,/Vj/, and/jV/stimuli, the boundaries were at stimulus numbers corresponding to frequencies of the relatively steady-state vowel-like segments. In experiment 2, the quadratic F2 trajectories of the/wV/stimuli were changed to linear trajectories and F1 and F3 trajectories were either quadratic, linear, or steady state. The results indicated that the shape of F1 and F3 trajectories had no effect on the boundaries. For the linear F2 trajectories, as for the quadratic F2 trajectories in the/wV/ stimuli, the boundaries were at lower stimulus numbers than for the stimulus with steady-state F2, indicating presence of perceptual compensation. In the experiment 3, the F1 and F3 were steady state and the linear F2 trajectories had three different values of frequency difference, delta F, between the initial and final frequencies. The perceptual compensation was found for the 200-ms stimuli with large delta F up to 500 Hz over 200-ms stimulus duration, and perceptual enhancement of final frequencies was found for the stimuli with delta F = 280 Hz.

Perceptual context effects of speech and nonspeech sounds: The role of auditory categories

“…Relative to the steady-state condition, more /I/ responses were obtained for the /wVw/ context and more / ʊ / responses were obtained for the /jVj/ context. Similar context-dependent vowel categorization has subsequently been demonstrated by Nearey (1989) using stop consonant contexts (/bVb/ vs /dVd/) and by Nabelek and Ovchinnikov (1997), using isolated vowels versus /wV/ contexts (with no steady-state portion).…”

“…The frequency of F2 increased in 10 Hz steps from 1000 to 2500 Hz. These endpoint values matched the stimuli used by Nabelek and Ovchinnikov (1997). Lindblom and Studdert-Kennedy (1967) and Williams (1986) varied their F3 frequency as a function of F2.…”

“…The four series were (1) isolated vowels synthesized to vary from /ʊ/ to /I/ (V series), (2) vowels synthesized in the /wV/ context /wV/ series), (3) sine-wave analogs with steady-state frequencies at the nominal formant frequencies for the isolated vowels (V-SW series; vowel-sine wave), and (4) sine-wave analogs of the wV speech series (wV-SW series). Based on previous work by Nabelek and Ovchinnikov (1997) and Williams (1986), it is predicted that the phonetic categorization of the speech series will result in a context effect, but that the categorization of the nonspeech series will elicit no context-moderated shift.…”

“…However, in the current study, the context condition corresponds to /wV/ as opposed to /wVw/. These stimuli are modeled in part after parameters used by Nabelek and Ovchinnikov (1997) in their extensive study of the acoustic parameters that affect the size of context effect shifts. They demonstrated that preceding the vowel with /w/ results in significant shifts in vowel categorization but following the vowel with /w/ does not.…”

Low Back Vowel Merger in Missouri Speech: Acoustic Description and Explanation

“…Carré et al (1994) manipulated synthesized formant patterns and found that perception of the vowel /a/ was affected only when the formant frequencies were far from canonical for this vowel. Nábělek and Ovchinnikov (1997) compared the perception of linear and quadratic vowel formants and found that a vowel identifi er categorized the differing vowel contours as the same. Pitermann (2000) concluded that steady-state portions of vowels alone serve better than dynamic aspects alone in modeling vowel perception in French, though this generalization may not hold for English.…”