Relationship Between Tasked Vocal Effort Levels and Measures of Vocal Intensity

Abstract

Purpose

Patients with voice problems commonly report increased vocal effort, regardless of the underlying pathophysiology. Previous studies investigating vocal effort and voice production have used a range of methods to quantify vocal effort. The goals of the current study were to use the Borg CR100 effort scale to (a) demonstrate the relation between vocal intensity or vocal level (dB) and tasked vocal effort goals and (b) investigate the repeated measure reliability of vocal level at tasked effort level goals.

Method

Three types of speech (automatic, read, and structured spontaneous) were elicited at four vocal effort level goals on the Borg CR100 scale (2, 13, 25, and 50) from 20 participants (10 females and 10 males).

Results

Participants' vocal level reliably changed approximately 5 dB between the elicited effort level goals; this difference was statistically significant and repeatable. Biological females produced a voice with consistently less intensity for a vocal effort level goal compared to biological males.

Conclusions

The results indicate the utility of the Borg CR100 in tracking effort in voice production that is repeatable with respect to vocal level (dB). Future research will investigate other metrics of voice production with the goal of understanding the mechanisms underlying vocal effort and the external environmental influences on the perception of vocal effort.

Voice and speech production involve a range of complex conscious and subconscious behaviors, including cognitive language formulation, as well as motor planning and execution of the respiratory, laryngeal, and oral–pharyngeal articulators. These components work in concert to adapt to changes in the environment or communication situation. One common adaptation is the Lombard effect, in which a person modifies their production (e.g., increased loudness, fundamental frequency, vowel modification, spectral tilt, and speech rate) to accommodate to background noise (Junqua, 1993; H. Lane & Tranel, 1971; Lombard, 1911). Other demands that result in vocal production adaptations include physical obstacles, such as distance (Liénard & Di Benedetto, 1999; Pelegrín-García et al., 2011) or poor room acoustics (Bottalico et al., 2015, 2016a; Rollins et al., 2019), as well as communication partner requirements, such as speaking to a child (Rowe, 2008) or to someone with a perceived hearing loss (Ferguson & Kewley-Port, 2007; Krause & Braida, 2004).

These vocal production adaptations usually require a change in the speaker's vocal exertion or vocal effort. While some individuals may not notice this change in effort, changes in vocal effort have been shown to be correlated with other vocal adaptations such as vowel modifications, along with changes in vocal fundamental frequency, dB SPL, spectral tilt, and speech rate (McKenna & Stepp, 2018). Unfortunately, if an individual does not notice when their effort increases, they may ignore early signals of vocal overuse (Whitling et al., 2017a, 2017b), and phonotrauma may result. On the other hand, although effort increases in those with a pathology (Cannito et al., 2012; Searl, 2020), increased vocal effort is at times an early indicator and may in fact be elevated even without remarkable changes in voice physiology or quality (Bermúdez De Alvear et al., 2011). This may explain why vocal effort ratings appear to strongly predict scores on the Voice Handicap Index (Eadie et al., 2010; van Leer & van Mersbergen, 2017). Furthermore, elevated vocal effort (Bach et al., 2005; Cannito et al., 2012) is the most commonly reported complaint (van Mersbergen et al., 2021) for many individuals presenting with voice disorders.

From both a speech science and clinical vocal health perspective, it is necessary to understand not only how and why vocal effort changes but also how it is perceived to change; both will clarify the relationship between the communication demands placed on a speaker, the necessary adaptations the speaker makes to these demands, and the potential effects of these adaptations on the speaker's vocal health (Berardi & Hunter, 2020). Hampering our efforts in this understanding are at least three critical barriers: (a) a consensus on the definition of vocal effort has not been universally accepted; (b) even though researchers and clinicians regularly assess vocal effort (regardless of which perspective they take), a standard, validated vocal effort measure has not yet been widely adopted; and (c) without a consensus definition or a standard measure, it has been difficult to isolate the underlying psychophysiological factors contributing to vocal effort.

Defining Vocal Effort

While previous studies have defined vocal effort as the perceived exertion to create voice, a consensus on the definition of vocal effort has not been universally accepted. Instead, vocal effort has been approached from at least two very different perspectives. First, speakers have been asked to rate their self-perceived exertion from the physical act of voice production (Halpern et al., 2009; Searl & Evitts, 2013; van Leer & van Mersbergen, 2017). Previous studies have quantified vocal effort in this way using a range of perceptual scales (e.g., Bottalico et al., 2015; Chang & Karnell, 2004; van Mersbergen et al., 2008; van Mersbergen & Delany, 2014). In each of these studies, there is both a physical event (voice production) and a perceptual phenomenon of exertion (vocal effort) from the perspective of the speaker. The second perspective that previous studies use to approach vocal effort is from the listener's judgment of the vocal effort heard in a voice (Eadie & Stepp, 2013; Eriksson & Traunmüller, 2002; Traunmüller & Eriksson, 2000) or the corresponding acoustic metrics of the produced voice (ISO 9921, 2003; Traunmüller & Eriksson, 2000). In some studies, these two factors were combined by relating acoustic quantities (e.g., vowel formants, fundamental frequency) to listener judgments of vocal effort (Liénard & Di Benedetto, 1999; Traunmüller & Eriksson, 2000). This auditory perception perspective is also embedded into vocal quality scales (e.g., GRBAS [grade, roughness, breathiness, asthenia, strain] scale, Hirano, 1981; American Speech-Language-Hearing Association's Consensus Auditory-Perceptual Evaluation of Voice, Kempster et al., 2009), in which a rater judges the degree of strain or exertion perceived in a speaker's voice.

Toward establishing a universal definition of vocal effort, a recent consensus paper reviewed how vocal effort (along with vocal fatigue, vocal load, and vocal loading) was used in the literature (Hunter et al., 2020). Furthermore, borrowing the definition of effort from the well-established fields of ergonomics and exercise science (i.e., the self-perceived perception of exertion in response to a demand; e.g., Ford-Baldner et al., 2015), vocal effort was delineated from similar terms to be the self-perceived exertion during vocalization (Hunter et al., 2020); consequently, the listener's judgment of the vocal effort heard in a voice or the corresponding acoustic metrics of the produced voice conceptualized as vocal strain. Thus, this article will approach oral communication from the following perspective: When a speaker has a need to communicate in a particular environment (vocal demand), there is (a) the physical act of voice production (vocal demand response), (b) the self-perception of the speaker (vocal effort), and (c) the perception of the listener (vocal strain).

Measuring Vocal Effort

The field of psychophysics, including the exploration of the relationship between physical and perceptual phenomena, was pioneered by the work of Weber (1850) and Fechner (1860) and resulted in the Weber–Fechner law that the relation between stimulus and sensation is logarithmic. Critical to communication sciences, forerunners such as Fletcher and Munson (1933) used psychoacoustic scales, most of which are nonlinear (e.g., sone and phon). Generally, these foundational psychoacoustic scales align with Steven's power law, which has superseded the logarithmic-based Weber–Fechner law with a power relationship (perceived magnitude increase of a perceptual response is proportional to the square of the stimulus) in order to describe more sensory comparisons even down to zero (Stevens, 1971). Underlying these relationships is that mental number scales seem to use nonlinear or compressed scaling in the coding of cognitive magnitude (Nieder & Miller, 2003).

Nevertheless, vocal effort has often been quantified with linear scales. For example, several vocal effort studies have used the Visual Analog Scale (VAS); this psychometric response scale with linear spacing does not need a physical stimulus or extensive training, but it does require strong anchors and may use valid midpoints (e.g., maximum effort, average effort, no effort). Studies using VAS have demonstrated a relationship between vocal demand, vocal effort, and listener and acoustic description of produced voice (e.g., Bottalico, 2017; Bottalico et al., 2015, 2016b; McKenna et al., 2019; McKenna & Stepp, 2018). These studies have added insight to our understanding of vocal effort; however, because the VAS can be difficult to interpret mathematically without external interpretation and manipulation, examiner bias is possible when categories are imposed onto the scale and/or mathematical error is possible because ordinal properties exist without interval guidelines or a 0 point (E. Borg, 2007; G. A. Borg, 1982; Ford-Baldner et al., 2015). Furthermore, because VAS scales are linear, they may lack the appropriate compressed scaling necessary to capture the intricacies inherent in psychophysical phenomenon and cognitive magnitude assessments.

A few voice studies have already used nonlinear perceptual scales. For example, two different studies (Chang & Karnell, 2004; Verdolini et al., 1994) measured vocal effort using a direct magnitude estimation (DME) scale, which requires a perceiver to indicate a numerical judgment of the sensory level of a known physical stimulus that can be manipulated to respond to the corresponding perception (e.g., 100 = a comfortable amount of effort during phonation, 200 = twice the comfortable effort, 50 = half the comfortable effort). In these studies, DME of perceived phonatory effort had some degree of success. However, the effectiveness of such a scale for vocal effort is reduced because DME scales need significant training to complete. Furthermore, and perhaps more importantly, DME scales assume there is a physical correlate that relates to the perceptual phenomena; however, because vocal effort is multifactorial, there is not a direct vocal effort physical correlate (even though it likely includes physical correlates).

Turning to other fields that have a history of examining the perception of effort from physical exertion, the most commonly used effort scales are nonlinear based on the substantial research spearheaded by Gunnar Borg for rating perceived exertion during physical activity (e.g., Dedering et al., 1999; Geddes et al., 2008; Hummel et al., 2005; Mahler et al., 2001; Yantis et al., 2002). A variety of Borg-based perceived effort scales have been successfully used to quantify perceived exertion or effort in areas such as sports, exercise, pulmonary function, and pain (Fanchini et al., 2016; Ries, 2005). A thorough review of the work of Gunnar Borg and the development of the Borg scales can be found elsewhere (e.g., E. Borg, 2007). Briefly, however, the Borg category–ratio (Borg CR) scale contains components of VAS and DME scales by combining the mathematical precision of a DME scale with the usability of the VAS (G. A. Borg, 1982; Ford-Baldner et al., 2015). Furthermore, the Borg CR scale has specified anchor and midpoint descriptions (e.g., slight, moderate, severe, and very severe) at nonlinear intervals to reflect the nonlinear relationship between the physical/physiological property in the perception of work and excursion.

One Borg scale, the Borg CR10, has been used in vocal effort studies. While the scale has been shown to be sensitive to changes in mood (van Mersbergen et al., 2008; van Mersbergen & Delany, 2014; van Mersbergen et al., 2017), cognitive status (van Mersbergen et al., 2019), and posttherapy outcomes (van Leer & van Mersbergen, 2017), the scale delineation (10 specified levels of effort) lacked sensitivity to be clinically meaningful (Ford-Baldner et al., 2015). To mitigate the lack of specificity of the Borg CR10, the Borg centiMax Scale or Borg CR100 (E. Borg, 2007) was developed to circumvent this problem by including midinteger values, thereby having the capacity to reflect changes that a 10-point scale may not capture.

In summary, speakers consciously or subconsciously adjust their vocal effort during voice production to meet a communication goal. Because vocal effort captures the perception of a speakers' exertion, it is important to test the reliability and repeatability of a tool that would quantify levels of vocal effort within an individual and, thereby, provide additional information about the processes of communication unique to the speaker. To that end, the purpose of this magnitude production study is to explore the repeatability of a vocal production level (dB) in response to a prescribed vocal effort level (VEL) goal in nondisordered vocal systems. Following the traditional approach of Stevens and others (e.g., Stevens, 1971), such relationships are approached with both magnitude production and magnitude estimation experiments. The specific research questions are as follows: (a) To what degree does voice production change as a function of a prescribed VEL goal? (b) Are the variations of voice productions consistently repeatable if a speaker is given the same VEL goal multiple times? As population characteristics may affect perception of effort and/or the resulting voice production, biological sex was included as a factor during the analysis based on the hypothesis that voice production and perceptions may differ based on the biological sex of a speaker (Hunter et al., 2011). Perceived vocal exertion will be quantified using the Borg CR100 scale with voice-related descriptors and recognizable reference cues for end anchors; voice production will be quantified in terms of a single speech production metric, vocal intensity, or level in dB. In this magnitude estimation study, other vocal production metrics (e.g., fundamental frequency, cepstral peak prominence, and spectral tilt), as well as questions related to differences due to various vocal populations (e.g., aging, training, limitations), will be left to future studies.

Method

Participants

Twenty-two college-age participants (11 women and 11 men; M _age = 20.4, SD = 2.2, range: 18–29) were recruited to participate in this study (Human Subjects Review Board Protocol #13-1149) through a research pool and scheduling system managed by the Department of Advertising and Public Relations at Michigan State University. After following consent procedures, the participants were screened for hearing capabilities of at least 20 dB HL (both ears with pure tone at 500, 1000, 2000, and 4000 Hz) and responded to questions related to vocal health, such as diet, sleep, reflux, vocal demands, and vocal health history (Cantor-Cutiva et al., 2018, 2020), as well as several published voice indices such as Vocal Handicap Index-10 (VHI-10), Vocal Fatigue Index (VFI), and 10-Item Big Five Inventory (BFI-10); Hunter & Banks, 2017; Nanjundeswaran et al., 2015; Rammstedt & John, 2007; Rosen et al., 2004). None of the 22 participants reported a history of voice problems, demonstrated noticeable voice production issues, nor had voice scores outside the normal range.

Recording Equipment

The study took place in a sound isolation booth (IAC, 2.5 × 2.1 × 2.0 m single-walled, 24 dBA background noise). The participants were fitted with a head-mounted microphone (omnidirectional; Countryman B3) placed approximately 5 cm to the side of each participant's mouth, with a second microphone (Behringer ECM8000) that was used for reference placed at 50 cm from the mouth just off axis to the front of the participant. Both microphones were connected to XLR inputs on a digital handheld recorder (ZOOM H5 Handy Recorder, 16-bit, 44.1 kHz, wav). The reference microphone was absolutely calibrated to 94 dB SPL (relative to 20 μPa) using a half-inch sound calibrator; the head-mounted microphone signal was calibrated to the reference microphone signal following a two-stage calibration procedure (Švec et al., 2005). The calibrated signals were adjusted to be SPL at 1 m (+6 dB). Data processing and analysis of the recordings are discussed below.

Speech Tasks at Cued Effort Levels

The experimental stimuli was presented using a laptop (Dell Inspiron 7472) with an open-source python-based experiment tool, PsychoPy (Peirce et al., 2019). During the experiment, the participants vocalized a presented stimulus (see Table 1, randomly ordered) at a cued VEL goal from the Borg CR100 scale (see Figure 1). Four VELs were chosen to represent a range of common vocal efforts (VEL 2: minimal vocal effort, VEL 13: slight vocal effort, VEL 25: moderate vocal effort, VEL 50: severe vocal effort). Three types of speech stimuli were used to represent a range of common speech patterns.

Table 1.

Three types of speech stimuli (automatic speech, read speech, and structured spontaneous speech), each with three variations.

Figure 1.

Borg CR100 scale of vocal effort. The vocal effort targets used in the study are noted. VEL = vocal effort level.

Each speech stimuli (see Table 1) was designed to be completed in about 15 s, at an average speaking rate of an English speaker. The first type of speech stimuli, automatic speech, prompted the participant to (a) say the alphabet, (b) count from 1 to 25, and (c) say the days of the week and the months of the year. The second type of speech stimuli, read speech, required the participant to read excerpts from three commonly used speech passages: (a) Marvin Williams Passage (Popolo et al., 2005), (b) an excerpt from the Rainbow Passage (Fairbanks, 1960), and (c) Stella Passage (Weinberger & Kunath, 2011). The final type of speech stimuli, structured spontaneous speech, tasked the participants with describing three different routes on a subway map, inspired by Anderson et al. (1991). The map was shown to the participants, along with audible instructions in the following format: At a vocal effort level of X, describe how to get from A to B via C.

Procedure

Prior to producing the speech tasks, the participants first engaged in a short tutorial to introduce them to the task, which allowed them to practice the vocal effort scale and each of the speech stimuli. To reduce uncertainty and to mitigate possible experiment–administration bias, tutorial and experimental instructions were given both by text on a computer screen and also by artificial speech narration (van den Oord et al., 2016). For example, while being shown the map, the participant would read text instructions (Describe how to get from Hillsboro to the Airport via Gateway.) and would hear oral instructions (If asked to describe how to get from Hillsboro to the Airport via Gateway you would say something like, “Starting at Hillsboro, I will take the blue line eastbound towards Beaverton. I will pass Beaverton, Pioneer Square, and the Rose Quarter. I will change at the Gateway station to the red line northbound to the Airport. And finally, I will arrive at the Airport.”) The participants then would be asked to practice describing a route on the map. The vocal effort scale was introduced with reference cues for end anchors for the extreme values of 1 and 100: 1 = quietly talking to someone next to me at home, 100 = trying to shout at someone while standing on an airport runway. The two reference cues for end anchors were chosen to elicit a sense of vocal effort range. For example, while an individual could speak quietly and comfortably at a VEL of 1, an individual could deliberately speak more quietly in a confidential voice, but this would potentially take more vocal effort to produce. After this initial introduction, only the scale (see Figure 1) was used to cue the participant to produce the requested vocal effort.

The participants were given the opportunity to take a short break before the collection of the speech tasks; they also were offered bottled water to drink as needed. During the speech task collection, the vocal effort scale was shown with a specific VEL highlighted visually and aurally like this: At a vocal effort level of 25, count from 1 to 25. The participants were instructed to complete each of the nine speech stimuli at each of the four VEL goals for a total of 36 trials. The order of presentation was randomized for each participant. While the presentation of each trial was monitored by the experimenter, each was self-paced, with the participant able to choose when they were ready to continue to the next trial. Most recorded samples were between 12 and 16 s of speech, with a few near 20 s; therefore, the total cumulative speech elicited from a participant was between approximately 8 and 12 min, with the total participation time from consent to conclusion about 30 min.

Data Processing and Analysis

Given a specific speech stimuli and a cued VEL goal, the primary speech production variable of a stimuli was the overall speech vocal level (SVL) calculated from the calibrated head-mounted microphone recording (absolute dB, ref 1 m, 94 dB tone). While other acoustic metrics could have been used (and can be the focus of future studies), the goal of this study was to test the repeatability of the vocal effort scale and not the various acoustic measures that would correlate with vocal effort. While the reference cues for end anchors that were chosen are not synonymous with intensity, they may be associated with intensity and were further chosen given that (a) previous research has indicated that vocal level (dB) is the primary acoustic parameter that changes with effort (ISO 9921, 2003; McKenna & Stepp, 2018) and (b) studies have indicated that self-perceived vocal level (autophonic response) is highly repeatable and correlated to actual production intensity produced (H. L. Lane et al., 1961; Yadav & Cabrera, 2017), with the rate of increase in the autophonic response slightly overestimating actual voice production level increases where listener judgment of vocal level increase significantly underestimates production level. The SVL from each recorded speech stimuli was obtained using the following three steps: (a) speech level for the whole stimuli was first estimated in 0.02-s signal intervals collected in 0.01-s steps, (b) nonvoiced intervals were disregarded using a voicing detection algorithm (custom MATLAB scripts combined with PRAAT), and (c) the overall average SVL from the remaining voice-only intervals was logged for the recorded stimuli. This procedure then provided a single SVL (in dB, ref 1 m) for each of the 36 recorded stimuli for each participant. Future studies may look more in detail at dB variance within a stimuli and other voice and speech features.

Statistics were performed using SPSS statistics software. To address the first research question (i.e., To what degree does voice production change as a function of a prescribed VEL?), SVLs of all recorded stimuli (used as a representation of voice production) were compared to the four cued VELs; SVLs within each of the cued VELs were checked for normality, independence, and equal variance. To determine whether there would be a systematic difference in SVL for a given VEL, a general linear model (GLM) with SVL as the dependent variable and cued VEL (2, 13, 25, 50) as an independent variable, with an alpha level of p = .05, was used. Post hoc Tukey's honestly significant difference tests were used to compare the speech production of the VEL pairs. Additionally, to provide insight into whether a population characteristic (i.e., biological sex) may affect perception of effort and/or the resulting voice production, self-reported biological sex (male or female) and its potential interaction with the cued VELs were included in the GLM. To address whether variations of voice productions would be consistently repeatable if a speaker were given the same cued VEL goal multiple times, a repeated measure reliability test was determined for each individual participant using Pearson r (which estimates the strength of a linear association). Outliers within the data set (greater than the sum of the third quartile and 1.5 times the interquartile range [IQR] or less than the difference of the first quartile and 1.5 times the IQR) were removed; contributing SVL samples after outlier removal are listed in Tables 2 and 3.

Table 2.

Descriptive statistics for speech vocal level (in dB at 1 m) for the cued vocal effort level (VEL) productions across all participants.

Voice level (dB)			95% Confidence interval for mean
	M	SD	Lower bound	Upper bound	Minimum	Maximum	N
VEL 2	58.26	6.16	57.32	59.20	44.37	71.18	167
VEL 13	62.98	6.58	62.01	63.95	52.05	78.27	178
VEL 25	66.59	6.98	65.57	67.62	54.33	87.94	180
VEL 50	73.24	7.24	72.18	74.31	56.01	89.17	180

Table 3.

Multiple comparison statistics for speech vocal level (SVL) for the cued vocal effort level (VEL).

(I) VEL	(J) VEL	Mean SVL Diff	SE	Sig.	Cohen's d	Pearson r	95% Confidence interval
		(I − J)					Lower bound	Upper bound
2 (N = 167)	13	−4.72	0.73	< .001	−0.741	−.347	−6.6	−2.85
	25	−8.34	0.73	< .001	−1.266	−.535	−10.21	−6.47
	50	−14.98	0.73	< .001	−2.228	−.744	−16.86	−13.11
13 (N = 178)	2	4.72	0.73	< .001	0.741	.347	2.85	6.6
	25	−3.62	0.72	< .001	−0.533	−.258	−5.46	−1.77
	50	−10.26	0.72	< .001	−1.483	−.596	−12.1	−8.42
25 (N = 180)	2	8.34	0.73	< .001	1.266	.535	6.47	10.21
	13	3.61	0.72	< .001	0.533	.258	1.77	5.46
	50	−6.65	0.71	< .001	−0.934	−.423	−8.48	−4.81
50 (N = 180)	2	14.98	0.73	< .001	2.228	.744	13.11	16.86
	13	10.26	0.72	< .001	1.483	.596	8.42	12.1
	25	6.65	0.71	< .001	0.935	.423	4.81	8.48

Results

Of the 22 participants, data from two were excluded (one opted to not complete the study and the other could not because of computer failure). For each of the cued VELs, the distributions of SVLs across trials and remaining participants (10 females and 10 males) were normally distributed and independent; equal variance could be assumed. Speech samples for analysis included a total of 720 elicited speech samples (9 stimuli × 4 VELs × 20 participants). Figure 2 shows an example of all SVLs from a female and a male participant plotted against cued VELs. The descriptive statistics of the SVLs are presented both in tabular and graphic form (see Figure 3 and Table 2). As would be expected, the SVLs generally increased with increased VELs when no other voice production requirements were given.

Figure 2.

Representative data from a male and a female participant, where the speech vocal level (dB) for each speech sample and vocal effort level (VEL) is plotted. The data for the male participant were offset to the right to better differentiate the two sets of data.

Figure 3.

Box plots for speech vocal level (dB) for the four cued vocal effort levels (VEL). The “×” represents the median, the thick lines represent the first quartile (Q1) and the third quartile (Q3), the whiskers represent the 1.5 × IQR above Q3 or below Q1. If the minimum or maximum is outside the whiskers, it is marked.

A three-way analysis of variance was conducted to compare the main effects and interactions for SVL, VEL (2, 13, 25, and 50), and biological sex (male or female). The analysis was significant, F(3, 704) = 84.94, p < .001; furthermore, within this model, the main effect of cued VEL was also significant, F(3, 704) = 169.89, p < .001. The post hoc tests (Tukey's honestly significant difference) showed that SVL was different across each VEL (summarized in Table 3). There was an increase (p < .001) in SVL of 4.72 dB from VEL 2 to VEL 13, an increase (p < .001) of 3.62 dB from VEL 13 to VEL 25, and an increase (p < .001) of 6.65 dB from VEL 25 to VEL 50.

Like the cued VEL, the main effect of biological sex was also significant, F(1, 704) = 82.46, p < .001, although the interaction between these factors was not significant, F(3, 704) = 0.72, p = .541. Figure 4 shows the relationship of SVLs for biological females and males for each VEL, while Table 4 lists the SVL differences between biological females and males across each VEL. Across all effort levels, the biological females produced 5.46 dB less (from the medians) than the males (differences of the means between 3 and 5 dB lower across all VELs). For example, the average median SVL of VEL 2 for the males and females was 60.60 dB and 55.74 dB, respectively, with a difference in the median of 4.86 dB. Additionally, the males were more homogeneous as a group in their median SVLs than the females, as can be seen by the smaller range (see Figure 4) and the negative difference in male-to-female standard deviation and IQR (see Table 4).

Figure 4.

Speech vocal level (dB) for the four cued vocal effort levels (VEL) for biological female and biological male participants. The bars represent 95% confidence interval.

Table 4.

Descriptive statistics of the ∆SVLs (difference of biological male and biological female mean speech vocal levels in dB).

∆SVL	VEL 2	VEL 13	VEL 25	VEL 50	Avg
∆Avg	3.81	3.55	5.12	5.05	4.38
∆Stdev	−2.50	−3.83	−3.53	−4.34	−3.55
∆Min	5.58	4.43	4.69	10.26	6.33
∆Q1	6.82	7.20	7.77	7.66	7.36
∆Median	4.86	5.69	7.58	6.71	6.21
∆Q3	1.94	3.26	5.29	4.37	3.72
∆Max	0.37	−0.83	5.18	−0.17	1.14
∆IQR	−4.88	−3.94	−2.49	−3.28	−3.65

Note. All differences were significant to p < .0001. Mean and median difference of the SVLs are bolded. Mean standard deviation difference (∆Stdev) and mean interquartile range difference (∆IQR) are italicized. VEL = vocal effort level; Avg = average; Q1 = first quartile; Q3 = third quartile.

Furthermore, with a given VEL goal, participants produced the nine different stimuli, which became the basis of the repeated measure reliability analysis. The analysis for all participants indicated repeatability, as the Pearson r within each participant was significant (p < .001). Across all participants, the averaged Pearson r was .90 for the nine different stimuli for each VEL level.

Discussion

The results of this experiment demonstrate the potential for the Borg CR100 scale being used to quantify vocal effort. To do this, the effort scale needed to show a correspondence with vocal production changes (Research Question 1), as well as reliability in repeated trials (Research Question 2). Using cued VELs from the scale, participants were able to produce levels of voice (SVL) that were distinct and repeatable.

Using the Borg CR100 vocal effort scale, the four VEL goals elicited four distinct levels (dB) of voice commonly observed in everyday communication situations. The results are in line with expectations from prior reports where there were significant vocal level increases (dB) across increasing vocal efforts (Cushing et al., 2011; Rosenthal et al., 2014; Skinner et al., 1997), as well as with the ISO acoustic definition for vocal effort for speech level at 1 m (ISO 9921, 2003). These results expand this understanding by suggesting that, given a perceptual goal (vocal effort), the corresponding physical output (voice production) appears to vary in a predictable and consistent manner when using the Borg CR100, a nonlinear psychometric scale. Thus, this finding suggests that the Borg CR100 psychometric scale potentially may be used to further explore the relationships between VEL and voice production.

Participants' repeated trials were shown to be strongly correlated, which implies good reliability of the scale. This finding supports the assumption that vocal performance in terms of SVL will be consistent within participants, which (a) presents researchers and clinicians with a reliable and repeatable scale that will provide consistency in measurement across differing experimental tasks and (b) enables confidence that changes in vocal effort measurements would reflect changes in perception of vocal work. This level of accuracy will allow for an improved understanding of the physical and physiological contributions to changes in voicing and disambiguate physical and perceptual aspects of voicing behavior.

Furthermore, this type of a reliable, repeatable scale of vocal effort would be beneficial in further investigations of the contributions of other voicing measures beyond intensity. Using the magnitude production method, additional acoustic measures such as fundamental frequency might vary differently than intensity, thus providing a picture of the unique contribution of a loudness goal (as was elicited in this study) versus a pitch goal to an individual's sense of effort. Additionally, this methodology could be expanded to other physiological measures, such as electromyography, when studying the contributions of muscular activation on the sense of effort. While more basic research needs to be performed to unequivocally establish this scale as a reasonable vocal effort scale, the clinical applications of such a scale are appealing. Because elevated vocal effort may be considered a symptom of a voice problem (Solomon, 2008), measuring changes in vocal effort during/after therapy would lend a valuable, patient-centered measure in the voice clinic. Furthermore, understanding the relation between vocal production (what one does) compared to vocal perception (what one feels) would provide insight into how changes in vocal behaviors influence changes in vocal perception.

Finally, a biological sex difference was found in vocal production, a reasonable finding due to inherent sex-based physiological differences. For example, biological males, on average, have larger aerodynamic and phonatory systems (Hunter et al., 2011; Kent, 1993; Titze, 1989) and subsequently appear to use less physiological engagement to reach the same vocal level than biological females (Hunter et al., 2019; Smith et al., 2019). Such a reduction in physiological engagement likely would be reflected in VELs. This finding may suggest that one of the factors for the higher prevalence of voice problems in biological females (when all else is equal) could be that biological females may need to use more vocal effort to achieve the same speech level (e.g., teaching in a noisy classroom could require more effort; Hunter et al., 2019; Smith et al., 2019).

Beyond the direct population difference in biological males and females, the results point to possible use when comparing other populations with physiological differences. Since the results of the GLM indicated that the interaction was not significant, the biological sex differences can be interpreted without considering the interaction effect; therefore, the magnitude of the male–female differences may be represented with a linear offset. Treating biological sex as a population difference with a linear offset may imply that another population with a physiological difference could also be comparable once the population-specific offset is known, thus potentially allowing for a wider use of the scale.

As with all studies, there are limitations in the current study. The purpose of this study was to show both the repeatability of voice production for a specific level of vocal effort and the uniqueness of voice production between the four designated levels of vocal effort; the purpose was not to specifically investigate how voice production was dependent on VEL or to demonstrate that SVLs were the only indicator of voice production. Therefore, while the goal SVL was chosen as one of many potential representative indicators of voice production, other voice production correlates could have been used to give insight into various vocal production dependencies on vocal effort changes. The choice of using SVL as a primary indicator was supported by a previous study of vocal effort; participants in the study were instructed specifically to maintain their vocal loudness while changing vocal effort to document that other metrics of production were sensitive to effort changes, but (even with this constraint) there was a strong and significant correlation between vocal effort and vocal loudness (McKenna & Stepp, 2018). Future studies could test the sensitivity of other acoustic vocal production metrics to VEL changes; for example, vocal production metrics such as fundamental frequency, cepstral peak prominence, or spectral tilt may give important indications into how effort affects production. Furthermore, future studies could also test the relationship between aerodynamic voice metrics and VELs given that previous studies have shown that measures such as subglottic pressure correlate with perceived phonatory effort; nevertheless, such studies must be conducted judiciously given that aerodynamic measures do not always translate well to running speech tasks and present increased measurement variability.

Next, the study is limited by having elicited only four levels of vocal effort, about half the available range on the Borg CR100 scale. The scale used here follows conventions from foundational psychophysical work, which has investigated the number of categories on scales and reliability of self-ratings (Bendig, 1953). Nevertheless, even with a limitation to VELs commonly found in everyday oral communication rather than what would be vocally possible, these four levels resulted in distinct vocal productions. Future studies will be needed to explore how vocal production changes with smaller VEL intervals, as well as with more extreme targeted VELs. Furthermore, the study may have also been limited by the instructions given to participants for using the vocal effort scale. While participants were provided with what were assumed to be identifiable reference cues for end anchors, the effectiveness of the scale would be reduced if a participant could not properly imagine the reference cues presented.

Finally, while the results between targeted vocal efforts levels and between biological genders were significantly different, this study may be limited by the population of vocally healthy, college-age adults; further studies should be expanded to other populations with physiological differences and limitations. Nevertheless, this study's techniques and results provide the groundwork for such future studies of vocal effort. Additionally, while this study's methodology required participants to produce voice at a prescribed VEL goal (magnitude production), future studies could look at the reverse relationship where participants would be put in different communication environments (e.g., quiet room, loud noise, speaking confidentially, speaking to a classroom of students) and asked to produce appropriate vocalizations and provide their vocal effort (magnitude estimation). Future research could also incorporate the Borg CR100 vocal effort scale to build on previous studies that investigated acoustic or physiological measures of vocal performance and vocal effort (e.g., Bottalico, 2017; Eriksson & Traunmüller, 2002; McKenna et al., 2019; Rosenthal et al., 2014), along with the effect of mood, cognitive load, and self-regulation (e.g., van Mersbergen et al., 2019; van Mersbergen & Delany, 2014).

Conclusions

Vocal effort, defined as the self-perceived exertion during vocalization, will vary with changes in accommodations to vocal demand. Thus, a reliable and validated measure of vocal effort is an important tool to study the nature of voicing, its environmental demands, and responses to these demands. The current study illustrated the relationship between produced vocal level and vocal effort goals when presenting vocal effort on a psychophysical scale, the Borg CR100. Participants reliably produced different speech levels for specific VELs with clear distinction and reliable repeatability. Additionally, differences in VEL and production between biological females and males further support that the scale is sensitive to physical and physiological voicing properties. Therefore, the Borg CR 100 appears to provide an improvement over the use of the Borg CR 10 scale, which reflected statistically significant but clinically indistinct effort levels (Ford-Baldner et al., 2015; van Leer & van Mersbergen, 2017).

Such a standardized protocol to assess psychometric VEL (patterned after the Borg rating of perceived exertion scale) could provide clinicians and researchers a meaningful outcome measure of vocal state. Partnered with other currently used tools, this scale could facilitate more holistic assessment and monitoring of vocal health state in clinical and research settings. Moreover, this tool could be used to train self-awareness of phonatory exertion, allowing for a straightforward means for an individual to articulate physiological experiences and providing the means to self-monitor behaviors by periodically self-tracking VELs.

Author Contributions

Eric J. Hunter: Conceptualization (Equal), Data curation (Supporting), Formal analysis (Supporting), Funding acquisition (Lead), Investigation (Equal), Methodology (Equal), Project administration (Lead), Resources (Lead), Software (Equal), Supervision (Lead), Writing – original draft (Equal), Writing – review & editing (Equal). Mark L. Berardi: Conceptualization (Supporting), Data curation (Lead), Formal analysis (Lead), Funding acquisition (Supporting), Methodology (Equal), Writing – original draft (Supporting), Writing – review & editing (Supporting). Miriam van Mersbergen: Conceptualization (Supporting), Methodology (Supporting), Writing – original draft (Supporting), Writing – review & editing (Supporting).

Funding Statement

The research reported in this publication was supported by National Institute on Deafness and Other Communication Disorders Award Number R01DC012315 (PI: E. J. Hunter). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.