GRBAS and CAPE-V

GRBAS is an auditory-perceptual metric that is used to subjectively assess voice based on five characteristics: grade, roughness, breathiness, asthenia, and strain. Grade assesses hoarseness and abnormality in the voice. Perceived roughness represents the degree of irregular vocal fold vibration. Breathiness assesses glottic air leakage [15, 16]. Asthenia measures perceived weakness in the voice. Strain is an assessment of perceived vocal hyperfunction. The five components of the GRBAS system are rated individually on a four-point scale, where zero corresponds to normal phonation and three corresponds to severely disordered phonation [17].

The Consensus Auditory-Perceptual Evaluation of Voice (CAPE-V) is another psychoacoustic metric that allows clinicians to assess voice based on overall severity, roughness, breathiness, strain, pitch, and loudness. This system uses a 100-mm-long visual analog scale, where the far left represents normal vocalization and the far right represents severely disordered voice [8, 18]. Both GRBAS and CAPE-V require a trained rater, typically a speech-language pathologist, to make judgements on a subject’s voice. Thus, as mentioned previously, these metrics are subject to the rater’s level of experience.

Both GRBAS and CAPE-V have been used with children to successfully distinguish between healthy and disordered voices and evaluate the outcomes of surgical procedures and voice therapy [19, 20, 21, 22, 23, 24]. The reliability of these measures has also been investigated. In a study of 50 children aged 4–20, the CAPE-V metric was used to assess dysphonia after laryngotracheal reconstruction. Seventeen of the samples were then rerated at a later time. Inter-rater reliability was high for perceptions of breathiness, roughness, pitch, and overall severity with intraclass correlation coefficients (ICC) ranging from 67% to 71%. Perceptions of loudness were less reliable (ICC = 57%). Except for strain, intra-rater reliability was strong for all parameters, with the ICC ranging from 63% to 93% [25]. Other studies in adults and children have also found that strain, measured using CAPE-V or GRBAS, is less reliable [26, 27, 28].

Fundamental Frequency

Fundamental frequency (F_o) is the lowest frequency of a periodic or nearly periodic signal. Within a complex periodic waveform, there exists a period (duration of a single cycle of a periodic signal) that is the smallest overall. This is the fundamental period T_o. F_o is then defined as 1/T_o [1]. F_o reflects the frequency of vocal fold vibration; qualitatively, changes in fundamental frequency will result in a change of pitch. This is calculated through a Fourier transform, which is a decomposition of the time-domain signal, in this case an acoustic recording, into its frequency components. Figure 8.1 depicts the different waves that can make up a signal. The frequency of each wave can be plotted along with their relative amplitudes. The lowest of these frequencies is taken as F₀.

In a study of 218 healthy children aged 4–17, fundamental frequency was measured from recordings of four CAPE-V sentences. For three of the four sentences, fundamental frequency was found to decrease significantly more rapidly during ages 11–14 for boys, compared to ages 4–11 and 14–17. For girls, fundamental frequency linearly decreased with age. There was no critical age at which frequency dropped more significantly [29].

Jitter and Shimmer

Jitter and shimmer are parameters that track perturbation in the voice. Specifically, jitter is a measure of the change in fundamental frequency from cycle to cycle, and shimmer measures cycle-to-cycle change in amplitude of the signal [1]. Jitter and shimmer have been used extensively in acoustic analysis; however, these methods are less reliable in the analysis of aperiodic voice signals [12, 30, 31, 32].

Signal-to-Noise and Harmonic-to-Noise Ratios

Signal-to-noise ratio (SNR) quantifies how dominant the voice signal is over random noise. Harmonic-to-noise ratio (HNR) is similar to SNR in that it quantifies the dominance of periodic (harmonic) signal elements over noise [33]; however, noise in this case refers to random noise, aperiodic signal elements, and perturbations such as jitter and shimmer [34]. This parameter is helpful for assessing vocal characteristics such as breathiness, which results from turbulent airflow [33]. Both SNR and HNR have been used to successfully analyze healthy and disordered voices in children [35, 33, 36, 37].

Perceptual and signal processing techniques can also be used in combination to provide a comprehensive acoustic analysis of voice. In a study of 39 children aged 7–14 with a diagnosis of bilateral vocal nodules, acoustic analyses were performed to assess progression during voice therapy. Therapy lasted for 8 weeks and consisted of lessons on vocal hygiene and voice abuse reduction, breathing and phonation coordination, and laryngeal massage. Jitter and shimmer improved most significantly post-therapy. HNR was also lower after therapy compared to baseline, but F_o did not change. Perceptual ratings according to the GRBAS system showed that grade, roughness, breathiness, and strain all improved [36].

Correlation Dimension, Lyapunov Exponents, and Kolmogorov Entropy

Nonlinear dynamic methods are useful for analyzing normal and disordered voices, and they provide the advantage of not having to track cycle boundaries or fundamental frequency, which can be difficult for aperiodic voices [38, 39, 40, 41]. Correlation dimension (D₂), Lyapunov exponents, and Kolmogorov entropy are nonlinear dynamic methods that have been used extensively in research [12, 13, 30, 42, 43, 44]. D₂ measures the number of degrees of freedom that are necessary to describe a dynamic system, with more complex systems having a higher D₂ [43]. In the context of voice analysis, D₂ objectively describes the degree of periodicity or aperiodicity and chaos in the voice. When D₂ does not converge to a finite value, this indicates the presence of a high level of random noise [43]. Lyapunov exponents assess a dynamic system by focusing on two trajectories that are initially nearby and measuring their rate of divergence or convergence over time [44]. For voice samples that are nearly periodic, the signal remains stable, and the value of a Lyapunov exponent remains close to zero; however, chaotic signals have positive Lyapunov exponents [30]. Second-order Kolmogorov entropy (K₂) quantifies the rate that information about the system dynamics is lost, with a positive K₂ value indicating chaos in the signal [43]. The calculation of Lyapunov exponents, D₂, and K₂ involves reconstructing the phase space of the corresponding signal, which is used to describe all possible dynamic states of the voice signal over time [30, 45].

Meredith et al. performed nonlinear dynamic acoustic analysis of sustained vowels in 23 dysphonic and 15 healthy children. D₂ was higher among dysphonic children, indicating that the voices of dysphonic children require a higher number of degrees of freedom to be fully quantified and are therefore more aperiodic. Additionally, though jitter was higher among dysphonic children, variability was high for both groups [46].

Linear and nonlinear acoustic analyses were also utilized in a study of 111 healthy female and 101 healthy male children aged 6–12. While jitter and shimmer did not vary significantly with age or sex, fundamental frequency and largest Lyapunov exponent were lower in boys and decreased with age. These findings show that the boys’ voices had lower frequency and were more stable than girls’ voices [47]. Higher Lyapunov exponents have also been seen in cleft palate patients with hypernasality compared to cleft palate patients without hypernasality [48].

Formants

Resonance created by vocal tract filtering can be described by formants, which are characterized by a center frequency and bandwidth and are influenced by the length and shape of the vocal tract [49]. Similar to fundamental frequency, changes to formant frequency and spacing may provide important information about the voice [49, 50, 51].

In a study of ten 5-year-old children with cerebral palsy, formant measures were used to investigate dysarthria [51]. Analysis of single word recordings showed that children with dysarthria had smaller second formant ranges when uttering words that are known to require larger changes in vocal tract shape [51]. Formants were also used to collect normative acoustic data in Brazilian children aged 4–8. Seven different vowels were uttered by each of the 207 children. Frequencies of the first three formants were generally higher in girls than boys, and formant frequency decreased with age [52].