Introduction
Vibrato regularity—defined as the cycle-to-cycle predictability of modulation—matters for technique, style, and pedagogy, yet has rarely been quantified at scale in commercial recordings due to historical barriers in vocal separation and quality control.
Methods
A corpus of 150 commercial recordings of G. F. Händel’s “Ombra mai fù,” spanning all classical voice types and career tiers (Bunch & Chapman, 2000), was processed with two independent separation pipelines: MusicAI and iZotope RX11 Advanced. Within each recording, six sustained notes were extracted, normalized, and analyzed in VibratoScope (batch mode). Exported data were evaluated in two phases: (1) separate computation of Coefficient of Variation (CV) and Sample Entropy (SampEn) at note and singer levels; (2) dynamical assessment via recurrence analysis on densely reconstructed f₀ trajectories, extracting Determinism (DET) and mean diagonal length (L). A principal component analysis (PCA) over (DET, L, SampEn) yielded a composite regularity index, with higher PC1 indicating longer, more deterministic recurrences and lower entropy.
Prior to statistics, a cycle-screening script labeled HasVibrato from VibratoScope’s cycle exports. TRUE required sufficient contiguous cycles with consistent period and amplitude; files lacking continuity were excluded. After screening, MusicAI and iZotope outputs were matched by filename, prioritizing MusicAI when both were valid. Singer IDs were mapped to voice type and career category. Group differences were tested using Kruskal–Wallis and pairwise Wilcoxon (Holm), complemented by unweighted/weighted linear models and balanced resampling to mitigate unequal group sizes.
Results
Recurrence plots indicated moderate-to-high DET and stable L, and PC1 loaded positively on L and DET and negatively on SampEn, coherently indexing regularity. By career category, Superstars displayed higher regularity than several other tiers; in weighted models, Superstar and National/Big City remained above Amateur, and International showed intermediate regularity with broader spread, including multiple least-regular cases. By voice type, Boy Sopranos tended to be least regular, while adult trebles (e.g., Soprano) and Tenor ranked among the most regular—especially for CV of extent and the DET/L composites. After sample-size balancing, the voice-type effect for CV of extent persisted. Singer-level Top/Bottom 10 derived from DET+L showed strong concordance with the PCA index.
Conclusions
Integrating dispersion (CV), complexity (SampEn), and temporal structure (DET, L) yields a scalable and musically meaningful representation of vibrato regularity in commercial recordings. Higher regularity in specific career tiers and voice types aligns with expectations of technique stabilization and fach-related constraints, offering practical value for pedagogy, performance, and clinical assessment. The study provides data-driven insight into vibrato regularity among highly professional singers who serve as pedagogical reference models.