Work

Electroglottographic and acoustic metrics of vocal technique in trained lyric sopranos: a descriptive cross-dataset study

Year: 2026
Type: article
Abstract: Vocal technique in lyric sopranos is a central topic in voice science and vocal pedagogy, yet existing studies are commonly limited by small samples, the lack of systematic physiological–acoustic comparison, and weak ecological validity. In this study, electroglottographic (EGG) and acoustic signals were recorded simultaneously from 24 trained lyric sopranos, yielding 1,104 valid tasks across three vowels and four pitch points for four techniques—vibrato, straight tone, breathy, and trill—and a cross-dataset comparison was performed within the acoustic feature space using the female subset of VocalSet (154 segments). Between-technique comparisons were descriptive, summarised as group means with 95% confidence intervals and effect sizes rather than by hypothesis testing. Breathy tasks were descriptively lower in the contact quotient at a 25% threshold (CQ 25 ), the dEGG closing-to-opening peak amplitude ratio (dEGG C/O ratio), cepstral peak prominence (CPP), and the harmonics-to-noise ratio (HNR), and higher in the first-to-second-harmonic amplitude difference (H1 − H2) and shimmer, whereas trill tasks were descriptively higher in EGG-jitter, the standard deviation of the fundamental frequency (F0 SD), modulation rate, and modulation extent than vibrato, an elevation consistent with intended F0 modulation rather than cycle irregularity. Within-subject Spearman correlations indicated consistent associations between EGG and acoustic measures, the strongest being between EGG-jitter and acoustic jitter ( r = 0.864 pooled across techniques; within-technique median r = 0.575). Random forest classification trained on the experimental data and tested on VocalSet reached an accuracy of 0.961 and a macro-averaged ROC-AUC of 0.966, whereas reverse transfer from VocalSet to the experimental data was lower (accuracy 0.889); trill was the least reliably recovered technique in the forward direction (recall 0.848). These results indicate asymmetric segment-level transfer across the two datasets examined, but do not establish singer-independent generalization because singer composition, recording conditions, equipment, and task format differed simultaneously. This study proposes, as a descriptive framework requiring further validation, a two-level EGG–acoustic metric structure covering glottal contact, closing dynamics, cycle stability, source-spectrum quality, and resonance focusing, which may serve as a candidate set of variables for the objective description of lyric soprano techniques and for cross-dataset research.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236660005
FWCI:
Cites: 33
Cited by:
Related to: 10