Quantifying Vocal Repertoire Tessituras Through Real-Time Measures

Abstract

Introduction:

Voice teachers use anecdotal evidence and experience in determining the appropriateness of repertoire for each student’s development. Tessitura is important in that determination, but until recently a straightforward, repeatable, and quantifiable method for determining tessitura has not existed. However, technology exists to provide the means to estimate the tessituras of standard vocal repertoire by measuring sung pitch (fundamental frequency) and vocal dose (amount of phonation) in real-time performance.

Purpose:

The purpose of this study was to examine the combined use of tessituragrams, Voice Range Profiles (VRPs), a singer’s self-perception of a performance, and expert listeners perception of a performance towards the goal of a more systematic way of selecting appropriate voice repertoire for singers. The following research questions guided this investigation: 1) How do Performance Range Profiles (PRPs, performance-based tessituragrams computed from neck skin surface vibration during singing), compare to score-based tessituragrams of the same selection in the same key? 2) How do PRPs of the same vocal score compare when performed in three different keys? 3) How do singer VRPs compare with PRPs of three performances of a score, each sung in a different key? and 4) How do singer and expert panel perceptions of the selection’s “fit” in three different keys compare with the alignment of each singer’s VRP to their PRPs? Four female singers and five expert voice pedagogues were enlisted to address these questions.

Results:

The distribution (histogram) of the score-based tessituragram closely matched the distribution of performance-based tessituragrams (PRP), suggesting that score-based tessituragrams have promise in accurately reflecting the performance-based tessitura of a musical vocal work. Acquired data revealed relationships of practical importance between score-derived tessituragrams, PRPs, acquired VRPs, and singer perceptions of ease in singing. These data suggest that score-based tessituragrams aligned with singer VRPs show promise in repertoire selection. However, there was only a minor relationship between expert panel perceptions of ease in singing and the acquired PRPs or singer perceptions of ease. Creation of a score-based tessituragram database could be combined with singer VRPs to assist in appropriate repertoire selection.

Introduction

Singing voice pedagogues use experience and anecdotal evidence when selecting appropriate repertoire for students. During the repertoire selection process, teachers make observations about a selection’s range and tessitura, passaggio points, perceived vocal weight and color, and various other pedagogical goals that make that selection a good “fit” for a singer. Among these considerations, tessitura is a factor that could be quantified via computer in a systematic and repeatable way.

The “tessiturogram,” a diagram of the distribution (number of occurrences or duration of time) of all pitches (measured by frequency in Hz) in a selection of vocal music, was first suggested in the late 1980s by Thurmer as an aid in the selection of vocal repertoire.¹ The idea was revisited and further explored by Titze,^{2, 3} who suggested that the acquisition of singer Voice Range Profiles (VRPs) combined with the quantification of repertoire tessituras could help voice teachers scientifically choose repertoire that is a good “fit” for individual voices.

The Voice Range Profile is a clinical voice assessment tool that measures the full range of a singer’s vocal intensity (in dB SPL) at each pitch in their vocal range (Figure 1). The use of VRPs could be helpful in choosing repertoire because teachers and students may not be adequately aware of students’ vocal capabilities and limits. Lycke, Decoster, and De Jong investigated voice classification practices among conservatory-based classical and musical theatre voice teachers. Finding no consensus on the need for voice classification, they instead found that when voice classification was practiced, no standard criteria were used to establish those classifications. Further, they suggested that the uncertainty around voice classification among teachers and students could represent a potential risk due to a lack of awareness which could in turn cause students to exceed the physiological limits of their voices.⁴ Lycke and Siupsinskiene acquired VRPs from both females enrolled in singing lessons (N=162) and non-singing females (N=67). They found that more advanced singing students demonstrated a greater fundamental frequency (sung note) range and greater intensity (dynamic or dB SPL) range than non-singers, but that register transitions as measured by the VRP were not affected by training or institution.⁵

Figure 1.

Example of a Voice Range Profile (VRP) acquired with a voice dosimeter. For each sung pitch, the full vocal intensity range is shown.

Past studies have found that in live performance tasks, singers may exceed dB SPL and frequency levels recorded the lab. For example, Emerich and colleagues compared laboratory acquired VRPs of actors with voice production on stage; they found that actors sometime exceeded the sound pressure levels in the mid-range of their VRP during expressive acting.⁶ In a study of 30 professional female opera soloists, Lamarche and colleagues also found that VRPs collected in a performance setting elicited higher percentages of phonation above 90 dB than a physiological VRP collected in a lab.⁷

Because of the differences in location (lab vs. stage) and the desire to not constrain the movements of a singer, capturing the voice in a systematic way can be difficult. The voice dosimeter was developed in the early 2000s to measure phonation activity in a real world, outside of the constraints and control of a laboratory, environment.^{8, 9} The voice dosimeter was preceded by early vocal accumulator studies with more recent research specifically tied to investigating voice disorders using ambulatory phonation monitoring.¹⁰ Voice dosimeters measure vocal dose, defined as vocal fold tissue exposure to vibration over time.^{11, 12} Rather than relying on acoustic audio recording methods, these devices used an accelerometer transducer to record skin vibrations in the neck. In this way, phonation activities are tracked in isolation from ambient sounds, allowing the tracking of voice use in real performance situations where there may be other performers or accompanying music. Vocal dose data collected by voice dosimeters typically include: 1) Phonation time dose (Dt) - the cumulative duration of time (hh:mm:ss) the vocal folds have actually touched in a given period; 2) Cycle dose (Dc) - the accumulated number of vibratory cycles (one complete sequence of opening and closing of the vocal folds) in a particular time period; and c) Distance Dose (Dd) - an estimate of “how far” vocal folds travel in a period of time incorporating total phonation time, fo, and amplitude into one dosage measure. Taken together, these measures provide a detailed picture of the volume and intensity of voice use. Multiple studies have used voice dosimeters to track and assess the vocal doses of singers in rehearsal and performance situations.^13–18

A limited number of studies have examined the tessitura of different works through the creation of score-based tessituragrams, employing Titze’s method of diagramming the distribution and duration of all pitches in a vocal music score with each semitone displayed one axis and the duration of each pitch on the other. Pizzorni, Schindler, Sozzi, Corbo, and Gilardone examined the tessitura of 67 roles in 14 of Verdi’s operas, and estimated Dt as well as the percentage of notes in different parts of the vocal range, including the passaggio. They suggested comparing a singer’s VRP to these score profiles in order to help determine the feasibility of these roles for individual singers.¹⁹

To date, a search of the literature resulted in only three studies that have examined the consistency between score-based tessituragrams, VRPs, and either PRPs or expert panel analyses of performance recordings. For example, Hanrahan created tessituragrams for Mozart arias and had five tenors complete VRPs;²⁰ each tenor was assigned one aria that aligned well with their VRP and one aria that did not align well. An expert panel found agreement with the alignments determined by the VRPs. Paolillo and Fussi compared tessituragrams of two female and two male operatic roles with singer VRPs and vocal dose data acquired during performances at Teatro alla Scala in Milan, Italy.^{21, 22} More than just the VRP, they included passaggio points in their score analysis and charted dynamic agility (the dB range, or loudest minus softest phonation available to a singer on each individual pitch) with the score-based tessituragrams to determine best fits for a singer. Their dosimeter analysis suggested that singers may be able to successfully complete a role that does not align with the portion of their range with the greatest dynamic agility, but with a greater cost in muscular effort and increased sense of vocal fatigue. Nix completed a pilot-study comparing a soprano’s VRP, score-based tessituragrams, and an audio recording analysis of tessitura in a Mozart aria.²³ He analyzed the aria recording to create a scatter plot of intensity range (db SPL) at each sung frequency (Hz) of the piece in the manner of a VRP and called it a Performance Range Profile (PRP). He suggested that VRP readings and score-based tessituragrams could be helpful in selecting repertoire for singing teachers when combined with PRPs created through automated voice dosimetry readings.

The purpose of this study was to build on these previous studies and examine the use of tessituragrams and Voice Range Profiles (VRPs) in selecting appropriate voice repertoire through both computer (signal) analysis and expert panel (perceptual) analyses. The following research questions guided this investigation:

1. How do Performance Range Profiles or PRPs (real-time tessituragrams computed from neck skin surface vibration) compare to score-derived tessituragrams of the same selection in the same key?;

2. How do PRPs of the same vocal selection (“Il mio bel foco…Quella fiamma” attr. Benedetto Marcello) compare when sung by four different females in three distinct keys?

3. How do singer VRPs compare with PRPs of three performances of this aria, each sung in a different key?; and

4. How do singer and expert panel perceptions of the selection’s “fit” in three different keys compare with the alignment of each singer’s VRP to their PRPs?

Methods and Procedures

The methods below include three major sections. The first describes the creation of a tessituragram from a musical score, the second describes the creation of a computerized tessituragram from a performance recording with a voice dosimeter (PRP, Performance Range Profile; individually contextualized by a singer’s capability as identified by a VRP), and the third describes the perception of tessitura by performers and listeners.

Tessituragram from score

For analysis, the standard Italian aria “Il mio bel foco…Quella fiamma che m’accende” was chosen because it possesses a wide vocal range and would be known to most potential participants. It is published in 5 keys (High: B-flat minor; Medium High: A minor; Medium: F minor; Medium Low: E minor; Low: D minor), allowing for relatively fine matching of a performance key with an individual’s vocal ability.²⁴ The aria’s minor key (as opposed to a major key) was not a consideration in its selection. Dose Time (Dt) was estimated by counting the number of beats on each pitch multiplied by the estimated beats per minute (bpm) in the traditional performance tempi. These tempi were estimated to be 55 bpm (quarter note beats) in the introductory recitative and 100 bpm (quarter note beats) in the main body of the aria. We also estimated Recovery Time (Rt) by counting the number of rest beats between the initial onset and final release of phonation, multiplied by the estimated quarter note beats per minute in the recitative and aria. This was inspired by a related technique used to track recovery time in schoolteachers.^25–27 Since singers breathe, use rubato, and observe tempo markings, we made the following adjustments to the score: Fermatas counted double the marked time (a standard performance practice), and we added an eighth rest for each unmarked breath that is traditionally observed. The estimated Dt distribution can be found in Figure 2.a.

Figure 2.

Score-based tessituragram. (a) Expected Dt by pitch in the key of Am. (b) Expected Dc by pitch in the key of Am. In this figure, fo is presented in half-step notes which also correspond to semitones (st, with 12 semitones per octave).

We also converted musical notes (e.g., Ab4) to frequency (e.g., 415 Hz) and multiplied the Dt estimates by frequency to estimate the Cycle Dose (Dc) distribution in each of the five keys presented in the 5-key Schirmer edition. This was important to our analysis because higher keys have a higher Dc for the same amount of phonation time than lower keys, and as such could potentially be fatiguing. As an example, Figure 2.b provides a graphical representation of the number of vibratory cycles at each pitch in the medium high key of Am at the estimated tempi.

Tessituragram from performance (PRP)

Participants

Four female singers (N=4) were selected as a sample of convenience. The study was approved by the Human Subjects Committee at the primary author’s university. Each singer completed a demographic profile including age, voice type, years of choral experience, years of voice lessons and history of vocal problems. None of the participants reported any history of vocal pathologies, and all four had previously sung the aria employed for the study. Participants included:

1. 17-year old soprano; college freshman voice performance major; 3 years voice lessons; 4 years choir

2. 18-year old soprano; college freshman voice performance major; 1 year of voice lessons; 13 years choral experience

3. 21-year-old mezzo-soprano; college senior vocal music education major; 3 years of voice lessons; 16 years choral experience

4. 37-year-old soprano; professional singer; 10 years of voice lessons; 10 years choral experience

Recording Equipment

As participants were to be recorded in both controlled lab environments (studio) and in a simulated performance environment with piano accompaniment (stage), it was imperative to have a uniform voice recording system that was also robust to room effects and ambient sound. Therefore, we employed a voice dosimeter setup (portable recorder and neck attached accelerometer) as described in detail in previous studies as there are no current commercial voice dosimeters available.²⁸ The neck accelerometer (Knowles BU-1771 Model accelerometer)²⁹ was housed in a collar of the VoxLog portable voice analyzer; and the collar was adjusted to comfortably fit the circumference of the participant’s neck (Figure 3). The accelerometer signal was recorded using a portable Roland R-05 digital audio recorder with storage to an SD card (stereo, wav format, 44,100hz, 16 bit, Figure 3). Participants carried the Roland R-05 in a Tune Belt Vertical Microphone Transmitter Carrier Belt worn around the participant’s midsection allowing freedom of motion during the collection time. While in the performance environment, a second and redundant recording was performed with the hall microphone placed at a fixed position in the center of the hall and recorded using a ZOOM H6 device (stereo XY microphone attachment, 90-degree angle, wav format, 44,100hz, 16 bit).

Figure 3.

(a) Dosimeter device, including a Roland RO-5 digital audio recorder and the collar of the Sonovox VoxLog dosimeter. (b) Sonovox VoxLog Collar with embedded accelerometer.

Recording the Voice Range Profile (VRP)

In the controlled lab environment, each participant completed a VRP task while being recorded with the voice dosimeter setup as described above. To complete the VRP, each participant sustained every pitch of their vocal range (from low to high) for 3–5 seconds on [ɑ]. They were asked to begin each note as softly as possible and crescendo to their loudest sound while adjusting registers as needed.” Pitches were cued from a piano tuned to A₄=440 Hz.

Performance Recordings

Participants sang the selected material in a university concert hall while standing in a marked position in the center of the stage. The participants performed with a live accompanist performing on an upright piano approximately three feet from the singer. The musical scores were positioned on a music stand placed in front of the singers’ torso and below the level of the mouth. Each singer sang the same aria (“Il mio bel foco”) three times in three different randomly ordered keys: their favored key, the adjacent higher key, and the adjacent lower key, all in the Schirmer 5-key edition. For instance, if the singer’s favored key was Medium High, they performed the Medium High, Medium, and High keys in random order. In order to observe how closely the score-based tessituragram estimates matched actual phonation activity in real time, the singer participants were instructed to sing the song as they would normally perform it, with their preferred tempi and use of rubato. The performances were recorded simultaneously with the voice dosimeter (worn by the participant) and a hall microphone described above.

Dosimeter Recording Analysis

Using GoldWave audio editing software (https://www.goldwave.com/), the sung performance as detected via accelerometer and recorded by the voice dosimeter was segmented into individual WAV files for the VRP and each of the three versions of the aria performances. Following the voice analysis protocol of a previous study,²⁸ a custom MATLAB script analyzed the acquired accelerometer signal, estimating fundamental frequency (fo, in cycles per second, Hz; or in semitone, st) and relative vocal intensity (dB) from voice-only parts of the singing.¹⁴ This estimation was done every 10 msec of the recording. Absolute dB SPL values were not necessary for this study as we were comparing dB values only within each participant, so we presented only the relative dB relationships. From the fo and dB data, voice dosimeter metrics were calculated³⁰ (time dose: Dt, Dc, and Dd).

Using these metrics (fo, dB, Dt, Dc, Dd) extracted from the performances allowed for illustrations and comparisons of multiple metrics for each key within an individual. First, a profile plot was created that showed the distribution of sung pitches over time, with time plotted against sung pitch; This created a performance-based tessituragram (Figure 4) to compare with the score-based tessituragram’s estimates (Figure 2). Second, dB and fo were used to create a Performance Range Profile (PRP) equivalent to the VRP plot³¹ for each performance (Figure 5). Figure 5a illustrates all of the individual 10msec segments of the sung aria in terms of relative intensity (dB) and sung pitch (st). In Figure 5b, the data is presented a multidimensional vocal intensity (dB SPL) and fo histograms (occurrences of 10 msec voicing segments) to illustrate where the singer spent most of the time on a particular pitch and vocal intensity. Next, treating the number of occurrences in the multidimensional histogram like an elevation, a contour map (concentric lines) of same elevation (number of occurrences) was created, where the inner circles indicated higher occurrences (amount of time) of a sung pitch and sound pressure levels (Figure 5c). Finally, from the contour map, a contour line within the PRP that captured 68.2% of all occurrences (68.2% of the most occurring singing of pitch and intensity, where 68.2% represents the +/− 1 standard deviation of all datapoints) which we considered the “tessitura” (Figure 5d). The area of this “tessitura” shape was listed in terms of an area of fo and intensity (st SPL). This process was repeated for each song performance and for the VRP as a point of comparison. A larger contour area indicated that a greater dynamic and/or pitch range was utilized in the performance. The 68.2% contour shapes of each song performance (PRP) were laid over each VRP to create a visual representation of how each singer used their voice in each performance as it related to their VRP (shown later in Figure 9 in the Results section).

Figure 4.

Dosimeter-derived tessituragram of “Il mio bel foco…Quella fiamma”

Figure 5.

Creation of the PRP from accelerometer data. (a) A scatterplot of vocal intensity (dB SPL) and fo (st). (b) A 3-D histogram plotting number of occurrences of intensity and fo. (c) Contour map created from the elevations of the above 3-D histogram (b) treating the number of dB and fo occurrences like an elevation; concentric lines of a uniform color represent elevation, where the inner circles indicate higher occurrences of pitches and sound pressure levels. (d) A replot of (a) with the blue contour enclosing 68.2% of dB and fo phonation occurrences during the song performance (+/−1 standard deviation); the area of the contour can be represented in terms of an area (semitone x dB).

Figure 9.

Singer Perception vs Panel Perception – Average of All Questions

Perception of Tessitura

Singer Perception Questionnaire

Singers completed a survey of their performance perceptions immediately following each individual song performance (Figure 6). They rated their performance based on five categories by making a tic mark on a Visual Analog Scale (VAS) with the left end of the scale being “Free/Efficient” and the right end of the scale labeled “Strained/Inefficient.” The categories included: 1) Overall ease in singing; 2) High notes; 3) Low notes; 4) Register transitions; and 5) Overall “weight” of the selection. Each printed VAS had a length of 100 mm and was measured in mm from left to right to provide a perceptual rating between 1 and 100, with 1 being the most free/efficient and 100 being the most strained/inefficient. Mean ratings were calculated for each question (among all 4 singers) and for each singer (the mean of all 5 questions rated by each individual singer).

Figure 6.

Example of the Visual Analog Scale (100mm) completed by singers and expert panel members after each song repetition.

Expert Panel Questionnaire

An expert panel (N=5) of voice pedagogues listened to all twelve song recordings in random order. Panel members were informed that they would listen to four singers singing three renditions, each in a different key, of the same aria all in random order. All five pedagogues held advanced degrees in voice and served as solo voice faculty at a US university. The unfiltered audio recordings from the hall microphone were uploaded to Google Drive and sent via e-mail to each listener. The recordings were labeled by number (1–12), and listeners were instructed to download the recordings and listen to them in order in a quiet location with quality speakers. Each panelist completed their listening using the same device and device settings for all selections. Upon listening to each selection, the panelists completed the same perceptual survey as described above.

Results

Tessituragrams

Score-based tessituragrams for “Il mio bel foco” both for Dt and Dc were created (Figure 2a and 2b respectively). The process was repeated for all five possible key areas. The matching real-time dosimeter outputs, with cumulated time on the y-axis and pitch on the x-axis, consistently matched the contour of the equivalent score-based tessituragrams despite the use of rubato and slightly different tempi. The expected time from initial phonation to final release based on the score analysis was 186.3s, with an expected Dt of 149.6s and an expected Rt of 37.2s. The actual mean recorded time was 196.4s with a mean Dt of 135.0 and a mean Rt of 61.4s. The expected ratio of Dt to Rt was 80.0% to 20.0%, while the mean recorded ratio of Dt to Rt was 68.7% to 31.3%. The actual and expected Dt and Rt for each performance as well as the mean totals are included in Table 1 and graphically represented in Figure 7.a. The expected vs recorded Dc for each performance are graphically represented in Figure 7.b.

Table 1.

Recorded vs Expected Dose Times and Recovery Times

	Recorded Dt (s)	Recorded Rt (s)	Recorded Total time (s)	Expected Dt (s)	Expected Rt (s)	Expected Total time (s)
17yo Sop H (Bm)	146.5	60.0	206.5	149.6	36.9	186.5
17yo Sop MH (Am)	146.4	60.1	206.5	149.6	36.9	186.5
17yo Sop M (Fm)	138.8	69.2	208.0	149.6	36.9	186.5
18yo Sop H (Bm)	121.0	78.0	199.0	149.6	36.9	186.5
18yo Sop MH (Am)	123.0	72.5	195.5	149.6	36.9	186.5
18yo Sop M (Fm)	126.1	69.4	195.5	149.6	36.9	186.5
21yo Mez MH Am)	140.7	61.3	202.0	149.6	36.9	186.5
21yo Mez M (Fm)	145.5	63.0	208.5	149.6	36.9	186.5
21yo Mez ML (Em)	148.0	62.0	210.0	149.6	36.9	186.5
37yo Sop H (Bm)	126.9	46.6	173.5	149.6	36.9	186.5
37yo Sop MH (Am)	127.3	46.7	174.0	149.6	36.9	186.5
37yo Sop M (Fm)	129.2	48.3	177.5	149.6	36.9	186.5
Mean - All Performances	135.0	61.4	196.4	149.6	36.9	186.5

Figure 7.

(a) Performance Dt and Rt – Score-based vs. dosimeter reading. (b) Cycle dose (Dc) of Performances – Score-Based vs. Dosimeter Reading.

VRP and PRP Relationships

The combined VRPs overlayed with the tessitura contours of all three performances for each singer are represented in Figure 8a–d with fo presented in terms of semitone to match standard musical notes. As found in previous studies, singers occasionally exceeded the upper SPL of their VRP at certain pitches in performance; nevertheless, the tessitura contours generally fell within the middle part of the singers’ pitch range while utilizing most of the dynamic range for the accompanying pitch.

Figure 8.

(a) Singer 1 – dB SPL VRP area overlayed with SRP areas. (b) Singer 2 – dB SPL VRP area overlayed with SRP areas. (c) Singer 3 – dB SPL VRP area overlayed with SRP area. (d) Singer 4 – SPL VRP overlayed with SRP Area.

Perceptual Ratings

Mean self-perception ratings for all questions from all four singers for the higher key, accustomed key, and lower key performances were compared to the mean ratings from the expert panel for the average ratings of all five questions (Figure 9). A comparison of Individual singer perceptual ratings and mean expert panel response ratings can be found in Figures 10 (Singer 1), 11 (Singer 2), 12 (Singer 3), and 13 (Singer 4). The inter-rater reliability of the expert panel was tested using Intraclass Correlation Coefficient/Cronbach’s Alpha, with a rating higher than 0.7 considered acceptable and 0.8 considered a good level of reliability (Table 2). This statistical test revealed weak overall inter-rater reliability, with only the ratings of high notes showing an acceptable level of consistency greater than 0.7.

Figure 10.

Singer 1 (17-yo Soprano) a) Self Perception; b) Expert Panel Perception. A higher number indicates less perceived ease in singing.

Table 2.

Expert Panel Inter-Rater Reliability

	Intraclass Correlation/Cronbach’s Alpha	95% Confidence Interval		F Test
		Lower Bound	Upper Bound	Value	df1	df2
Overall Ease of singing	.612	.108	.873	2.579	11	44
High Notes	.736	.391	.913	3.782	11	44
Low Notes	.392	−.400	.800	1.644	11	44
Register Transitions	.239	−.752	.750	1.314	11	44
Weight	.598	.075	.868	2.487	11	44
Average	.473	.228	.658	1.898	59	236

Discussion

VRP, PRP and Tessituragrams

The acquired data are instructive when comparing multidimensional consistency between score-derived tessituragrams, singer VRPs, singer PRPs, and perceptual ratings for each singer. The score-based tessituragram aligned well with the PRPs, with the expected pitch distribution of Dts for all participants closely matching the Dt pitch distribution anticipated by the score analysis. This suggests that a score-based analysis in creating tessituragrams could have a practical application because it may accurately represent the tessitura of a musical work. Additional work with a greater number of singers is needed in order to generalize these implications.

This said, the measured dose time was lower and recovery time was higher than initially expected based on analysis of the score, despite the fact that we accounted for traditional unmarked breaths. There could be several reasons for this discrepancy. First, singers perform with words, and as a result elicit unvoiced consonants. Time spent articulating unvoiced consonants is time during which the vocal folds are not vibrating and certainly decreased the overall phonation time. Additional studies could examine a large body of repertoire to determine a standard percentage of unvoiced phonation time in different languages. Another reason for the discrepancy could be in the way the singers breathed. “Il mio bel foco” is an aria in which performance practice requires a considerable amount of expression and rubato. Anecdotally, the tradition of rubato in classical vocal music suggests that the singer moves more quickly through portions of phrases and tapers the tempo at the ends of phrases. This expressive tempo leaves more time for a breath and vocal recovery between phrases, and good singers take advantage of this stylistic element. Finally, the voice recording and subsequent analysis may not have accounted for all phonation from the recording (due to thresholding and other analysis techniques in identifying voicing segments), but this is likely to be minimal since the distribution of frequencies was similar to the expected distribution.

The method of statistically obtaining the combined dB/pitch contours of the recorded song performances and laying them over the VRP created a clear visual representation of the relationship between the singers’ VRPs and their song performances. The tessitura contour overlays revealed that singers were able to obtain greater overall volume and dynamic agility (the difference between the loudest and softest phonation) in keys that aligned most closely with the VRP pitches for which the singers had the most dynamic agility. Likewise, the singers’ self-perception of ease in singing appeared to align with the acquired dosimeter-derived PRP/VRP contour overlays.

Singer Specific Discussion

Singer 1’s VRP displayed an evident passaggio area just above middle C in which she had less ability to sing loudly. This area aligned most closely with the lowest of her three key areas, and that this key had a lower overall dB SPL (Figure 8.a) than the upper two keys. This singer rated the lower key as perceptually the most difficult, particularly in terms of register transitions (81 out of 100) and weight (87 out of 100) with an average response of 45.6 out of 100 for all questions regarding the lowest key. By comparison, her average response to all questions in the highest key, which aligned most closely with the range of her VRP displaying the most dynamic agility, was 11.6 out of 100.

Singer 2 did not have an obvious passaggio area that presented difficulty, but she was able to sing with more volume and had the widest dynamic agility on the pitches that corresponded most closely with the upper two keys. She rated overall ease, register transitions and weight least free and efficient in the lowest key. She rated the highest key most free and efficient in terms of low notes, register transitions and weight, but scored the high notes at 94 out of 100. It should be noted that as “Il mio bel foco” only briefly touches on the highest pitches, those pitches are not included in the PRP contour. The upper pitch in the highest key was Bb5, which was in the area of her VRP where she was not able to sing as loudly or with as much dynamic agility.

Singer 3, a mezzo-soprano, found the higher key the most difficult, with an average perceptual score of 52.1, compared to an average score of 12.0 for her accustomed key and an average perceptual score of 15.2 for the lower key. In the lower two keys, she exceeded upper dB threshold of her VRP in the lower pitches of the song, which corresponds with other studies in which singers were easily able to exceed their upper VRP limits when singing expressively. The upper pitches of the highest key correspond with an area of her voice that had less dynamic agility and less ability to sing softly, which corresponds with her sensation of less ease and efficiency in this key.

Singer 4’s VRP revealed a small but noteworthy passaggio point in the middle of her vocal range. The tessitura contour of the lowest of her three key areas lay over the middle of this weak area, whereas the upper two keys lay mostly above it. This singer had a much larger dynamic area in the upper two keys (128.8 stSPL in the high key and 132.7 SPL in the accustomed key) than the lower key (77.3stSPL). This singer’s perception corresponded with the visual analysis of the VRP and PRP. Her average score on for the lower key was 53.5, compared to 30.8 for the accustomed key and 23.6 for the lower key. She particularly found the low notes in the low key to be more strained and inefficient with a score of 77.

Expert Panel Perception

The Expert Panel Perception showed little inter-rater reliability or alignment with singer perceptions or VRP/PRP comparisons, which is not uncommon in perceptual voice analysis.^{32, 33} Whereas the singers indicated clear differences in sensations between different keys, the average difference in panel perception between the keys was small and did not clearly align with the VRP contours like the singer’s perceptions did. This suggests that singer self-perception of a vocal work’s comfort or fit may be more reliable than an expert pedagogue’s perception in selecting repertoire with an appropriate tessitura. This lack of perceptual difference in listeners is consistent with other studies that have demonstrated that individuals have the ability to create perceptually excellent vocal quality even when their voices feel fatigued or strained.^{14, 34–37} These corresponds to the distinctions of vocal production definitions recently described:^{37, 38} [1] vocal load, or vocal demand, is the vocal requirement of a situation and is independent of the vocalist, which in the case of this paper was the performance of the three arias and the VRP task; [2] vocal loading, or vocal demand response, is the physical act of voice production by a vocalist in response to the perceived vocal demand, which in this case was how the vocalist performed the tasks and is dependent on training, physiology, and other related characteristics; [3] vocal effort is the vocalists self-perception of vocal exertion during a vocal tasks, which in this case would be related to the perception of singer ease; and [4] vocal strain is the perception of the listener in judging the vocal effort of a vocalist. By considering these four parts as distinct terms though related, it is not surprising that the vocalists’ perception of the performance (vocal effort) is more related to the performance than the expert panels judgment (vocal strain).

Limitations and Future Directions

As with all studies, there are several limitations and options for future work. Due to the small number of participants and small expert panel, these results cannot be generalized but rather offer a demonstration of the technique and a point of departure for further study. Further, limitations related to automatic audio analysis may skew the results where a small amount of phonation activity may have been discarded at the onset or offset of voicing or conversely may have interpreted a small amount of vibrations from the piano accompaniment as phonation; never the less, these uncertainties would be applied the same to all singers and situations. Limitations such as these are discussed in more detail in a recent study by Bottalico, Ipsaro, Passione, Astolfi, Carullo, and Hunter.³⁹

A unique aspect of this study is that the VRP and song performances were captured by dosimeter technology, which measures phonation activity by capturing skin vibrations on the surface of the neck. This process means that pressure level readings captured the intensity of source vocal fold contact rather than filtered acoustic dB SPL readings. While previous studies have shown that skin vibrations can be calibrated to acoustic radiation,⁴⁰ it is uncertain how skin vibration relates to the perception of voice production (ease or effort) or the perception of a sung voice acoustically received (expert rating). More investigation on the relationship between neck surface vibration measurement (dosimeter) and acoustic measurement (audio) is needed.

Regarding the singer and expert perceptual ratings, while the visual analog scale allowed for better assessment of multi-dimensional features of voice quality, an argument can be made that such scales are better applied when raters have been trained in their use and allowed to practice with sample ratings. By not allowing our raters to practice, we may have introduced uncertainty in their results. Additionally, the labels we used to identify categories may not have aligned with performer or expert panel internal experience or definitions. The definitions of singing quality labels are not standardized in the profession of vocal pedagogy. Next steps could more rigorously investigate what raters perceive (including the examinations of labels, definitions, and categories of perceived singing quality), and how to define those perceptions.

Future studies could include singers performing in an unaccompanied situation which would also provide pure singing voice (without the piano or hall reverberation) for rating. However, our purpose was to simulate a performance environment (e.g., hall, accompaniment) as singers use their voices differently in a performance environment. Additionally, there are certainly many other factors that relate to tessitura that were not analyzed in this study, some of which are subjective or relate to traditional standards of performance practice – vowel structure, the distribution off ascending versus descending pitches, relative weight, etc. These considerations are described in detail in Nix’s pilot study.²³

Conclusion

There were relationships of practical importance between singers’ Voice Range Profiles, Performance Range Profiles, and singers’ self-perception of ease in singing a classical aria in three different keys. Key areas in which a greater range of pitches corresponded with areas of greater dynamic agility in singers’ VRPs tended to be perceived as more free and efficient by the singers. There was little relationship between expert panel perceptions of ease in singing and the produced voice metrics or singer perceptions, and there was weak inter-rater reliability among the panelists, suggesting that these singers may have been more self-aware of the best “fit” of a song in terms of tessituras than expert pedagogues. While the score based tessituragrams predicted more phonation time and less recovery time than the PRP dosimeter readings of the song performances, the frequency distribution of Dt readings closely matched. This suggests that score-based tessituragrams aligned with singer VRPs show promise in repertoire selection. Creation of a score-based tessituragram database could be combined with singer VRPs to assist voice teachers in appropriate repertoire selection.

Figure 11.

Singer 2 (18yo Soprano) a) Self Perception; b) Expert Panel Perception. A higher number indicates less perceived ease in singing.

Figure 12.

Singer 3 (21ys Mezzo) a) Self Perception; b) Expert Panel Perception. A higher number indicates less perceived ease in singing.

Figure 13.

Singer 4 (37yo Soprano) a) Self Perception; b) Expert Panel Perception. A higher number indicates less perceived ease in singing.