Evaluating the spectral distinction between sibilant fricatives through a speaker-centered approach

Abstract

This study was designed to examine the feasibility of using the spectral mean and/or spectral skewness to distinguish between alveolar and palato-alveolar fricatives produced by individual adult speakers of English. Five male and five female speaker participants produced 100 CVC words with an initial consonant /s/ or /ʃ/. The spectral mean and skewness were derived every 10 milliseconds throughout the fricative segments and plotted for all productions. Distinctions were examined for each speaker through visual inspection of these time history plots and statistical comparisons were completed for analysis windows centered 50 ms after the onset of the fricative segment. The results showed significant differences between the alveolar and palato-alveolar fricatives for both the mean and skewness values. However, there was considerable inter-speaker overlap, limiting the utility of the measures to evaluate the adequacy of the phonetic distinction. When the focus shifted to individual speakers rather than average group performance, only the spectral mean distinguished consistently between the two phonetic categories. The robustness of the distinction suggests that intra-speaker overlap in spectral mean between prevocalic /s/ and /ʃ/ targets may be indicative of abnormal fricative production and a useful measure for clinical applications.

Spectral dynamics of sibilant fricatives are contrastive and language specific

“…When estimated from either the beginning, middle, or end of the fricative noise, centroid frequency has been found to be greater in /s/ than in /S/ (Jongman et al, 2000), and similar results have been found when centroid frequency measurements from multiple locations across the fricative are pooled together Maniwa et al, 2009). In each of these studies, the difference in centroid frequency was evaluated across a group of talkers, and when the perspective shifts to the individual talker, the same pattern obtains (Haley et al, 2010;Li et al, 2009). As is the case with centroid frequency, peak frequency has been found to be higher in /s/ than in /S/, regardless of whether it is measured at the beginning, middle, or end of frication (Behrens and Blumstein, 1988;Heinz and Stevens, 1961;Hughes and Halle, 1956).…”

Associating the origin and spread of sound change using agent-based modelling applied to /s/-retraction in English

“…A discrete cosine transformation (DCT) was then applied to each speaker-normalized M 1 trajectory. The reason for representing M 1 trajectories with DCT coefficients is because of the evidence that sibilants in different contexts vary not just at the target but also dynamically (Haley et al 2010;Iskarous, Shadle, & Proctor 2011;Koenig et al 2013). The first three DCT coefficients are proportional respectively to the mean, linear slope, and curvature of the trajectory and thereby encode dynamic information between the acoustic onset and offset of each (speaker-normalised) M 1 trajectory (Watson & Harrington 1999).…”

Estimating the Acoustic Variation of s via Principal Component Analysis

“…В отличие от многих типов звуков для сибилянтов чрезвычайно трудно выделить такие акустические параметры, которые надежно отличали бы их друг от друга на уровне языка [Newman et al, 2001;Haley et al, 2010]. Кроме того, иногда сделать обобщения о качестве s в языке оказывается невозможным, потому что данные носителей слишком различаются [Gordon et al, 2002].…”

“…В лингвистической среде даже возникло мнение, что для сибилянтов вовсе невозможно выделить акустический инвариант, а есть только относительное расположение на спектральной картине (см. обзор: [Haley et al, 2010]).…”