From onset to coda: spectral variation in normative Polish /s/ produced by children

TL;DR — This pilot study investigates word-position effects on the spectral properties of perceptually normative Polish /s/ produced by 102 preschool children, finding significant positional variations driven primarily by initial-medial contrasts across spectral centroid, flux, kurtosis, and skewness.

Key contributions

• Quantified word-position acoustic variation (initial, medial, final) on 816 tokens of normative Polish /s/ from 5-8 year old children.
• Demonstrated that word-initial /s/ exhibits significantly lower spectral centroid, flux, and kurtosis compared to word-medial /s/ (Cohen's d up to 1.09).
• Revealed that medial-final contrasts are largely non-significant except for spectral skewness, positioning word-final /s/ as intermediate.
• Employed linear mixed-effects (LME) models controlling for stress, preceding/following phones, and parent word as random effects.

Problem

Sibilant fricatives like Polish /s/ are shaped by linguistic and developmental factors, yet word-position effects remain underdescribed in children's speech. Prior approaches often overlook how prosodic organization, boundary-related timing, and coarticulatory demands alter spectral properties even in perceptually normative productions. This gap makes it difficult to fully understand child speech motor development and establish language-specific acoustic baselines.

Method

The corpus consists of 816 tokens from 102 Polish-speaking children aged 5-8 years pronouncing 8 basic vocabulary words containing /s/ in initial, medial, and final positions. Audio was recorded via Panasonic WM61 electret microphones at 15 cm distance, sampled at 44.1 kHz with 16-bit resolution. Signals were manually segmented using 20 ms Hamming windows with a 10 ms overlap and 50 Hz spectral resolution STFT.

Five acoustic features were extracted per frame and averaged per token: first spectral moment / spectral centroid (CoG), spectral kurtosis (M4), spectral skewness (M3), and spectral flux. Linear mixed-effects (LME) models were fitted with word position as a fixed effect (initial as intercept), by-speaker random intercepts and slopes for stress, preceding phone, and following phone, and a random intercept for parent word. Pairwise contrasts were evaluated using estimated marginal means (EMMs) with Bonferroni adjustments.

Experimental setup

The dataset comprised 816 /s/ tokens from 102 preschoolers from the PAVSig database, spanning 8 words. Statistical analyses were performed in R 4.4.2 using lme4, with signal processing in Matlab 2023b. Metrics included estimated marginal means of CoG, flux, kurtosis, and skewness, evaluated via Bonferroni-adjusted pairwise contrasts and Cohen's d effect sizes.

Results

The initial-medial contrast yielded the largest effect sizes across all measures (d = 0.65-1.09, p < 0.001), with word-initial /s/ showing lower CoG, flux, and kurtosis than medial tokens. Initial-final differences were also significant (d = 0.47-0.56, p < 0.001), while medial-final differences were non-significant except for spectral skewness (d = 0.53, p = 0.02).

Limitations

The study relies on a small, unbalanced lexical corpus of only 8 words, limiting generalizability. Age was treated merely as a sample descriptor rather than modeled as a covariate, and no adult control group was included for baseline comparison.

Why read this

Phoneticians and speech researchers studying child language acquisition will learn how word boundary contexts systematically bias spectral moments of sibilants even in perceptually correct speech.

Code

None released (as of this page's updated date). If you are an author with a repo, please claim this entry — see CONTRIBUTING.md.

Applications

Speech-language pathology assessment tools, children's speech recognition systems, and developmental phonetics research.

Institutions

Silesian University of Technology

Funding / 經費: National Science Centre, Poland, European Funds for Silesia, Just Transition Fund

Related

• Age-dependent acoustic changes of the frication noise in dental sibilants produced by typically developing Polish children between 5 and 8 years of age — same problem · relatedness 2.1/3
• Detection of Incorrect Place of Articulation in Polish Sibilants Using Convolutional Autoencoders — complementary · relatedness 2.0/3
• Informativity of high-frequency bands on the place of articulation shift in retroflex sibilants produced by children — same problem · relatedness 1.9/3
• Sorting Clusters into the Shape of the Word: Positional Distribution of Probabilities — same problem · relatedness 1.9/3
• Phoneme-Level Mispronunciation Screening in Polish-Speaking Children with an Explainable Assistant — complementary · relatedness 1.9/3

All 950k paper pairs scored by TypeSafe Jev (scripts/related/); relatedness 0 = unrelated … 3 = directly comparable.