Introduction
Vibrato, an essential characteristic of vocal expression, has historically been studied in controlled laboratory settings, limiting research to small samples and often excluding elite performers. AI-powered voice separation now enables large-scale analysis of commercial recordings, providing access to vibrato characteristics across renowned singers, both living and deceased. This study applies these advancements to examine vibrato in professional performance contexts, connecting pedagogical insight with real-world vocal practice.
Methods
We analyzed 150 commercial recordings of G. F. Händel’s aria “Ombra mai fu,” performed by world-class, international, and national singers of all voice classes, classified using the Bunch & Chapman (2000) taxonomy. The orchestral accompaniment was removed using “music.ai” and “Izotope RX10,” approaches previously verified in a study with synthesized ground-truth samples. Within each recording, six sustained notes were selected for analysis. Vibrato rate and extent were computed using both the sine wave fit method (Nestorova et al., 2023) and a marked peaks method applied to fundamental frequency (f₀) contours; the two methods were combined to ensure robustness. All results underwent visual inspection through spectrogram analysis and a goodness-of-fit check.
Results
Clear trends in vibrato characteristics emerged across voice types and professional classifications. Higher voices, such as Altos and Sopranos, showed faster vibrato rates, with Altos reaching a median of 6.0 Hz, while Baritones and Basses exhibited slower rates around 5.4 Hz. Lower voices tended to have broader vibrato extents, while Boy Sopranos demonstrated the narrowest extents. Regarding professional taxonomy, “Superstar” and “International” singers displayed higher vibrato rates and narrower, more controlled extents, reflecting advanced vocal technique. In contrast, Local Community and Child singers showed broader extents and lower rates, suggesting less refined control. A multinomial logistic regression confirmed significant associations between professional level and vibrato characteristics, with “Superstar” and “International” singers consistently demonstrating higher rates and narrower extents, while broader extents and lower rates among Local Community and Child singers indicated distinctions in vocal maturity and technical development.
Discussion
This study represents the first large-scale investigation using AI-derived vocal separation to analyze vibrato in commercial recordings. The findings reinforce established understandings in vocal pedagogy, demonstrating distinct vibrato patterns across voice type and professional level. Higher vibrato rates and narrower extents in Superstar and International singers indicate refined control linked to advanced training and extensive performance experience. By providing quantitative insight into vibrato in real-world contexts, this research offers benchmarks for pedagogical evaluation and further study of its expressive role. The methods allow large-scale vocal analysis beyond laboratory limits and support continued exploration across diverse repertoires and singing styles.