Volume 53, Issue 4

Previous research has noted a reduction in perceptual identification performance when the speaker varies from stimulus to stimulus and has interpreted this finding as an effect of a normalization process that compensates for variability in the physical content of the speech signal. The purpose of the present investigation was to examine whether the physical variability introduced by this experimental design can result in general stimulus uncertainty effects that extend beyond the realm of traditional delineations of normalization processes. The stimuli were short segments taken from nasal consonant + vowel syllables produced by 1 male adult, 1 female adult, and 2 children. Segments of 25 and 50 ms duration were edited from the nasal murmur and the onset of the vowel. The stimuli were ordered according to four variability conditions and presented to listeners for place of articulation identification. The results showed that identification was significantly reduced when variability affected either speaker identity or segment type. Further analyses revealed that segment variability impaired perception of all segment types approximately equally, but that the 50-ms vowel segments were selectively spared in the speaker variability condition. These findings indicate that general uncertainty effects should be considered in speech perception experiments, and that dynamic properties of speech are particularly important in the perceptual compensation for speaker variability.

Acoustic correlates of stress [duration, fundamental frequency (F₀), and intensity] were investigated in a language (Thai) in which both duration and F₀ are employed to signal lexical contrasts. Stimuli consisted of 25 pairs of segmentally/tonally identical, syntactically ambiguous sentences. The first member of each sentence pair contained a two-syllable noun-verb sequence exhibiting a strong-strong (– –) stress pattern, the second member a two-syllable noun compound exhibiting a weak-strong (_˘ –)stress pattern. Measures were taken of five prosodic dimensions of the rhyme portion of the target syllable: duration, average F₀, F₀ standard deviation, average intensity, and intensity standard deviation. Results of linear regression indicated that duration is the predominant cue in signaling the distinction between stressed and unstressed syllables in Thai. Discriminant analysis showed a stress classification accuracy rate of over 99%. Findings are discussed in relation to the varying roles that F₀, intensity, and duration have in different languages given their phonological structure.

Measurements were made of the fundamental frequency of a register two-tone language, Kono, a Northern Mande language from the Niger-Congo group. This language has the tonal feature of downstep which conveys linguistic information in specific contexts and introduces a terracing effect on successive high tones. The physical values were converted into their musical scale equivalents and comparisons were made with a three-tone system. They were found to be different. Observations of local and global tone ranges were made and were discussed in terms of tone spacing.