New objective timbre parameters for classification of voice type and fach in professional opera singers

Abstract

Voice timbre is defined as sound color independent of pitch and volume, based on a broad frequency band between 2 and 4 kHz. Since there are no specific timbre parameters, previous studies have come to the very general conclusion that the center frequencies of the singer's formants are somewhat higher in the higher voice types than in the lower ones. For specification, a database was created containing 1723 sound examples of various voice types. The energy distribution in the frequency bands of the singer's formants was extracted for quantitative analysis. When the energy distribution function reached 50%, the corresponding absolute frequency in Hz was defined as Frequency of Half Energy (FHE). This new parameter quantifies the timbre of a singing voice as a concrete measure, independent of fundamental frequency, vowel color and volume. The database allows assigning FHE means ± SD as characteristic or comparative values for sopranos (3092 ± 284 Hz), tenors (2705 ± 221 Hz), baritones (2454 ± 206 Hz) and basses (2384 ± 164 Hz). In addition to vibrato, specific timbre parameters provide another valuable feature in vocal pedagogy for classification of voice type and fach according to the lyric or dramatic character of the voice.

Figure 1

Comparative presentation of two FHE values for the dramatic tenor (T1) Plácido Domingo (2632 Hz) and the lyric baritone (B2) Mario Cassi (3013 Hz) at comparable pitch f4 (362 vs. 361 Hz). The yellow highlighted area marks the frequency band defined for the calculation of all timbre parameters. The vibrato is calculated at the harmonic marked in red (VR, VE). The lower half of both examples shows the progression of sound energy within the investigated frequency band. The black bar indicates the frequency at which 50% of the energy is reached (FHE).

Figure 2

Correlation heat map showing the relationships between candidates for the input features in a random forest model. FHE frequency of half energy, Jitt Jitter, PHE position of half energy, Shim Shimmer, StaFo Start of formant, StoFo Stop of formant, StreFo Strength of formant, SC spectral centroid, SK spectral kurtosis, SS spectral skewness, SV spectral variance, VE vibrato extent, VR vibrato rate.

Figure 3

Importance of different features for voice structure classification of the investigated sopranos, tenors, and baritones (balanced error rates: 18.0%, 19.6%, and 17.5% respectively), based on information gain.

Figure 4

Left side: SHAP summary plots of input features for sopranos (upper row), as well as tenors and baritones (lower row). Features are sorted in descending order of the impact on the model output. Right side: SHAP dependence plots showing the impact of the interaction between SC and PHE. In all plots, only the results for the dramatic voice structure are shown.