Spectral dynamics of sibilant fricatives are contrastive and language specific

Abstract

Previous research has extensively investigated the spectral properties of sibilant fricatives with little consideration to how these properties vary over time. To investigate such spectro-temporal variation, productions of English /s/ and /S/ and of Japanese /s/ and /ˆ/ in word-initial, prevocalic position were elicited from adult native speakers. The spectral dynamics of these productions were analyzed in terms of a psychoacoustic measure of peak frequency: "peak ERB N number." Peak ERB N number was computed at 17 evenly spaced points across each fricative production. The resulting peak ERB N number trajectories were analyzed with orthogonal polynomial growth-curve models, to determine how peak frequency varied temporally within each fricative. Three analyses compared (1) the English sibilants to each other, (2) the Japanese sibilants to each other, and (3) English /s/ to Japanese /s/. The results indicated that, in both English and Japanese, the sibilant fricatives differ acoustically in terms of both static (i.e., overall level) and dynamic (i.e., shape) aspects of the peak ERB N number trajectories. Furthermore, English /s/ and Japanese /s/ exhibited language-specific differences in the shape, but not overall level, of peak ERB N number trajectories.

Acoustic–phonetic and auditory mechanisms of adaptation in the perception of sibilant fricatives

“…Some constraint must be in place to ensure that listeners match like with like when adapting to certain phonetic or auditory dimensions. In contrast, if the proper cue is not COG, but rather other aspects of the spectral shape including dynamic properties of the spectrum (e.g., Reidy, 2015Reidy, , 2016, it is certainly the case that the class of sibilants have more spectral similarities to one another than to the larger class of fricatives (specifically, sibilants have a strong spectral peak and an overall high amplitude). Moreover, the aspects of the spectrum that are perceptually extracted must extend beyond phonetic-specific cues given that listeners adapted in a highly comparable manner to both linguistic and nonlinguistic exposure stimuli.…”

Nasal Air Emission in Sibilant Fricatives of Cleft Lip and Palate Speech

“…This may deviate the temporal variation of the spectral properties (spectral dynamics) of fricative when it is produced with NAE. A state-of-the-art method to analyze the spectral dynamics of the unvoiced sibilant fricative is explored [12,17].…”

“…Therefore, the frame-rate varies depending on the speech sample. Then, the DFT magnitude spectrum computed from each segment is passed through a fourth-order gammatone auditory filterbank of 361 filters [17]. Then, the spectral energy at the output of each filter is summed up (termed as "auditory excitation"), and this auditory excitation represents the spectral envelope for a speech frame of a sound unit.…”

Methods for eliciting, annotating, and analyzing databases for child speech development

“…While sibilant fricatives can be modeled synthetically as being produced by static articulatory postures, recent research indicates that the spectral properties of sibilant fricatives do vary over time (Iskarous et al, 2011; Yu, 2016) and that the spectral-kinematic properties of sibilant fricatives carry language- and consonant-specific acoustic information (Reidy, 2016). Preliminary analyses of cross-sectional age differences in how the /s/ vs. / ∫ / contrast develops in preschool-aged children acquiring American English indicate development along both static and time-varying spectral properties (Reidy, 2015).…”

“…For the databases used in the two studies by Reidy (i.e., Reidy, 2015, 2016), these multiple tokens were elicited using the picture-prompted word-repetition task described in Edwards and Beckman (2008b). This task is an efficient way to elicit a reasonably large sample of productions of a number of target sounds.…”