Performing

lieder: Expert perspectives and comparison of vibrato and

singer’s formant with opera singers

Lynette Johnson-Read1, Anthony Chmiel1, Emery Schubert1, and Joe Wolfe2

1
Empirical Musicology Laboratory, School of the Arts and Media, University of New

South Wales, Sydney, Australia

2
Acoustics Group, School of Physics, University of New South Wales, Sydney,

Australia

Correspondence: e.schubert@unsw.edu.au

NOTE:

This is an accepted manuscript (pre-print version) of the below article. This paper is

not the copy of record and may not exactly replicate the authoritative document

Johnson-Read, L., Chmiel, A., Schubert, E., & Wolfe, J. (2015). Performing lieder:
Expert perspectives and comparison of vibrato and singer’s formant with opera
singers. Journal of Voice, 29(5), 645.e15-645.e32.
DOI: 10.1016/j.jvoice.2014.10.020
Performing Lieder: Expert Perspectives
and Comparison of Vibrato and Singer’s
Formant With Opera Singers
*Lynette Johnson-Read, *Anthony Chmiel, *Emery Schubert, and †Joe Wolfe, *ySydney, New South Wales, Australia

Summary: This article reports three studies about performance of lieder, and in particular in comparison with opera
performance. In study 1, 21 participants with experience in music performance and teaching completed a survey con-
cerning various characteristics of lieder performance. The results showed that there was consensus between the liter-
ature and the assessment of an expert panel—that a ‘‘natural’’ and ‘‘unoperatic’’ vibrato was favored, and that diction,
text, and variation of tone are all important aspects of lieder performance. Two acoustic analyses were conducted to
investigate genre-specific differences of the singer’s formant and vibrato parameters. The first analysis (study 2) used
18 single quasi-unaccompanied notes from commercial recordings of two lieder, and, for comparison, 20 single
unaccompanied notes from an opera. Vibrato rate was statistically identical between the two genres at !6.4 Hz; how-
ever, lieder featured a longer delay in vibrato onset. Vibrato extent was smaller for lieder (!112 cents) compared with
opera (!138 cents). The singer’s formant, which is generally associated with opera, was at times observed in the
lieder recordings; however, this was at an overall significantly weaker intensity than in the opera recordings. The
results were replicated in study 3, where recordings using only singers who performed in both lied and opera
were analyzed. This direct comparison used 45 lieder notes and 55 opera notes and also investigated three different
methods of analyzing the singer’s formant. A number of consistencies and inconsistencies were identified between
acoustic parameters reported in studies 2 and 3, and the beliefs of singing teachers and scholars in the literature
and study 1.
Key Words: Vibrato rate–Vibrato extent–Vibrato onset–Singer’s formant–Singing–Lieder–Opera–Expert panel.

INTRODUCTION Other genres of vocal performance have received more atten-

Teaching and learning to sing in different musical styles involve tion in recent studies. Research in the areas of opera and music
a capacity to execute characteristics that nuance the singer’s theater are more plentiful, examining text intelligibility,3
output so that the requirements of that style can be met. For vibrato,4,5 formant control,6 and other vocal characteristics.7,8
example, Sataloff1 states that vibrato can be produced in a Similar topics are also examined under the generic label of
different fashion for each genre of the singing voice, and that classical singing.9–20 Prame’s studies21,22 use F. Schubert’s
its presence is one factor which determines the character of Ave Maria (an Art song, though not strictly a lied) to examine
each genre. Our broad interest was to investigate whether a vibrato because it has many sustained notes throughout the
set of principles for all genres and styles of singing could be piece. Sustained notes provide greater scope for vibrato
available to reflect what performers do and to work toward effects, as will be discussed below.
well-defined guidelines to aid teachers and students. We were In Singing in style: a guide to vocal performance prac-
also interested in gathering perspectives from singing teachers tices, 23 Elliott discusses elements concerned with the
and performance and comparing those, where possible, with singing of lieder. On vibrato and portamento, she writes
acoustic measurement. ‘‘Vibrato can be used as an expressive device in lieder and
This article specifically looks at lieder, the 19th century is sometimes indicated with a wavy line or a variety of ac-
German form of song that is usually accompanied by piano. cent markings. A wide, continuous vibrato, however, is still
Through a variety of mediums, certain vocal techniques are inappropriate. Portamento, on the other hand, can be used
identified as being important for the performance of lieder. liberally in many situations for both small and large ges-
Vibrato and the singer’s formant (henceforth, SF–a prominent tures. It should be employed for expressive and dramatic
peak in the spectral envelope at around 3 kHz)2 are parameters purposes’’ (p. 192). The performance of lied is also dis-
that are specifically studied in voice acoustics. We also explored cussed in German lieder in the nineteenth century24 in which
the use of a qualitatively ‘‘pure tone’’ in lieder. Spillman refers to diction, vowel formation, consonants,
elisions, vocal ‘‘line,’’ and the pianist. He notes that ‘‘[t]he
singer must have a good ear for phonetic sounds and the
Accepted for publication October 28, 2014.
From the *Empirical Musicology Group, School of the Arts and Media, The University
technique and willingness to reproduce them faithfully .
of New South Wales, Sydney, New South Wales, Australia; and the yAcoustics Group, [as they are] frequently asked to do things that go against
School of Physics, The University of New South Wales, Sydney, New South Wales,
Australia.
basic bel canto training’’ (p. 318). Miller’s publication
Address correspondence and reprint requests to Emery Schubert, Empirical Musicology Singing Schumann: an interpretive guide for performers25
Group, School of the Arts and Media, University of New South Wales, Sydney, New South
Wales 2052, Australia. E-mail: e.schubert@unsw.edu.au
lists examples of how lied is sung poorly. Some inappro-
Journal of Voice, Vol. -, No. -, pp. 1-18 priate techniques are identified: ‘‘scooping into important
0892-1997/$36.00
! 2015 The Voice Foundation
words . starting the vocal tone straight ‘and then letting
http://dx.doi.org/10.1016/j.jvoice.2014.10.020 it wiggle’ with vibrato . removing vibrancy on notes of
2 Journal of Voice, Vol. -, No. -, 2015

short duration, changing the dynamic intensity of each note 3. What should a singer do to produce good lieder? [Please
in a phrase, using exaggerated ‘vocal colouration’ and ‘word provide as much detail as possible. Answer the question
painting’ to the detriment of vocal timbre’’ (p. 16). Although in every way that you can.]
these texts state how lieder should or should not be sung, the 4. What should a lieder singer NOT do? [Again, please
ideas are mentioned briefly and no specific explanations or provide as much detail as you can.]
demonstrations are provided explicating how singers can 5. [If not already mentioned] How would you say vibrato
use the techniques in practical situations. Nor do they apply should be used?
specific principles of vocal acoustics, which can be valuable 6. Do you think there is the need for the voice to have a
in documenting how lieder has, and can be, performed. No pure tone quality?
sources were found that made explicit statements about the 7. With respect to the previous question, what is your un-
use of SF in lieder, and reference to vibrato is mostly qual- derstanding of a pure tone quality?
itative even though vibrato is quantitatively well defined. 8. Do you have a recording of a good lieder performance?
However, such an approach to understanding lieder was If so, please provide the details of the recordings and
considered important in communicating the essence of lieder CDs (up to the four best). Please provide as much infor-
performance. mation as possible (including catalogue number) so that
We therefore examined some of these issues by conducting I can locate the recording.
three studies: one survey and two acoustic analyses of sound 9. Select one of the recordings from the above question.
recordings. The general aims were to investigate views on lie- Why is the performance good?
der singing through a sample of experienced and expert musi- 10. Is there anything you would change about the perfor-
cians (study 1) and then to follow this up by examining mance mentioned above?
selected acoustic parameters of lieder sound recordings in 11. Do you have a recording of lieder that you view as con-
comparison with opera (study 2). The acoustic analysis was taining a poor performance? If so, please provide the de-
then refined to an examination of singers who have made re- tails of the recording and CD. Please provide as much
cordings of lieder, but also in the comparison genre of opera information as possible (including catalogue number)
(study 3). so that I can locate the recording.
12. If you have mentioned a specific recording in the above
question, what do you think makes this poorly sung?
STUDY 1: SURVEY OF MUSICIAN’S BELIEFS ABOUT 13. What would you recommend that this singer change?
LIEDER PERFORMANCE How would you explain this to the singer?
The aim of study 1 was to investigate the perspectives among 14. Please provide information as to your musical back-
experienced teachers and performers of lieder, with regard to ground. Have you undertaken singing or music training,
performance of the genre. what experience do you have in singing and in the field
of lieder?
Methods
Participants. The recruitment of participants involved Results
sending e-mails to a range of music organizations within The comments from the survey were content analyzed to
Australia, including eisteddfods, examination boards, univer- pinpoint the techniques the participants of the survey identified
sities, opera and other vocal companies, lieder and art song so- as necessary for lieder performance, and whether there was a
cieties, and various contacts. Around 80 people were contacted, general consensus throughout. Each comment was put into a
of whom 21 completed the survey. Ethics approval was granted category. These categories included some that examined the
for the study. form of the answer, such as metaphorical, adjectival, or a tech-
From the recruited group, only three participants had prior nical explanation. Other categories described what the
contact with the authors. All but two of the participants had comment was in reference to, including items from Wapnick
vocal training. One of the participants had a postgraduate edu- and Ekholm’s27 criteria. Specifically, these categories were
cation in music, and another was a professional accompanist vibrato, diction (being the pronunciation of vowels and conso-
specializing in lieder and vocal performance. Ten participants nants), other translation issues (such as knowing the language
had completed a tertiary degree in music, and nine participants and exposing subtleties, as well as literal translations of the
described themselves as singing teachers. Several participants text), text interpretation, references made to ‘‘words,’’ compar-
had continued their vocal and instrumental training or perfor- ison with other genres, and/or piano accompaniment.
mance careers overseas. The following summarizes the content analysis results ques-
Stimuli and procedures. An online survey was prepared tion by question. Unless otherwise stated, all of the survey par-
using SurveyMonkey.26 The online survey consisted of 15 ticipants answered each question.
open-ended questions. These were as follows: Question 1: ‘‘What is your understanding of ‘lieder’?’’ The
first question was solely present to separate any respondents
1. What is your understanding of ‘‘lieder’’? without proper musical knowledge of lieder from other partic-
2. How would you describe a good lieder technique? ipants. All but four of the participants answered the question by
Lynette Johnson-Read, et al Performing Lieder 3

referring to the word ‘‘German,’’ and only two did not mention Questions 5: ‘‘How would you say that vibrato should be
the words ‘‘song,’’ ‘‘Art Song,’’ or ‘‘melody.’’ These participants used?’’ The general consensus (11 participants) was that a nat-
described lieder in a more conceptual style. The first wrote that ural use of vibrato was preferred. A moderate use of vibrato was
lieder is ‘‘A partnership between poem, accompaniment, and mentioned by seven participants although it was not clear what
vocal line,’’ and the second participant wrote that lieder is the difference was, if any, between a moderate and natural
‘‘Generally Poetry in German set to music. However sometimes vibrato. Eight participants specifically discouraged the
but rarely, poetry from other countries in other languages, set to maximal use of vibrato in lieder, stating that this was an unde-
music, is accepted in this genre.’’ All of the responses provided sired operatic sound. Respondents advocated the varied use of
sufficient information to confirm an understanding lieder. vibrato to enhance emotion (4), and one mentioned that the
Question 2: ‘‘How would you describe a good lieder tech- reason lieder did not need an exaggerated use of vibrato was
nique?’’ Knowledge and importance of text were emphasized because the singer did not have to make their voice heard
by eight participants. Good, clear diction, a centered sound, over an orchestra or in a large space; that the intimacy of lieder
with variations of tone color, good breath control and support, prescribed a more subtle use of vibrato. An additional possible
and vocal flexibility were other typical comments. Operatic acoustic implication of this belief is that lieder singers need not
style was reported by three participants as a comparison with use SF to the same extent, or not at all, compared with the opera
lieder, and this was often associated with an overdramatic vocal singer.2,27 One participant summarized the responses with the
quality if applied to lieder. Technical responses occurred in all comment that ‘‘[a]ll mature voices have vibrato which should
but four responses, and the possession of a good general vocal be controlled and used to effect when appropriate.’’
technique was mentioned throughout, such as ‘‘good breath Questions 6 and 7 asked the participants what their under-
support,’’ ‘‘a beautiful tone,’’ ‘‘[vocal] flexibility and control.’’ standing of a pure tone quality was. A typical response was
In comparison, four participants used metaphors within their re- ‘‘a focused sound, little perceptible vibrato, ie, a narrow and
sponses, such as ‘‘the sound has to be essentially sitting very even vibrato that gives vibrancy rather than intensity, a sound
high in the mask.’’ Some examples fell into multiple coding cat- with a ringing quality to get a ‘choir boy/early music’ sound
egories and were coded once for each occurrence, eg, ‘‘their without airyness.’’ One participant used more adjectival de-
voice maintained an ‘audible line’ [metaphor], well balanced scriptions, writing that ‘‘[t]he range of vibrato and tone quali-
in tone throughout the range [technical], displaying sensitivity ties is wider in lieder, but still a pure or ‘white’ quality
and color [adjectival], without an overriding vibrato [tech- should only be one of a vast range of colors and should be
nical].’’ Comments concerning the importance of accurately used very sparingly as it is very close to simply being poor tech-
conveying the meaning of the text through the voice were nique.’’ The responses to these questions suggest that purity of
mentioned by 16 participants and consideration of the piano ap- tone is a quality that is largely subsumed by the issue of vibrato.
peared in five responses. In combination with Question 5 and current literature, it also
Question 3: ‘‘What should a singer do to produce good lie- suggests consistent views on whether lieder should use wider
der?’’ aimed to generate responses describing how the ideal or narrower vibrato extent than other vocal forms, such as op-
techniques could be achieved by singers. Comments similar era. As a comparison, consider the suggestion by Elliott dis-
to the previous question were found here, but the responses cussed in the introduction. A precise use of the term vibrato
were explained in further detail and with a more personal focus. extent was not reported in the survey responses.
There was a greater use of metaphorical (7) and adjectival (8) Question 8: ‘‘Do you have a recording of a good lieder per-
responses, and specific teaching techniques were mentioned. formance?’’ The performers who were highlighted in good re-
The following explanation is an example of one with a technical cordings of lieder varied a great deal. Included in these
function: ‘‘vocal technique to produce a clear, pure, controlled choices were non-classical singers, such as Barbara Streisand.
sound.’’ Another participant wrote that ‘‘lieder requires the The performers who were mentioned more than once are shown
singer to have great breath control, limited vibrato, even tonal in Table 1. Of the pieces mentioned, the most frequently
quality throughout the range (no ‘breaks’ or ‘holes’), linguistic
proficiency, poetic interpretation, and a power over their audi-
ence—the ability to captivate the listener without the grandeur
TABLE 1.
of an operatic scene.’’
Performers Mentioned More Than Once in Question 8,
Question 4: ‘‘What should a lieder singer not do?’’ The par-
Pertaining to Good Performances of Lieder
ticipants wrote that one should not oversing (seven respon-
dents) or sing in the styles of music theater, opera, or other Performer Times Mentioned
genres (9). Participants also noted that the language should be Dietrich Fischer-Dieskau 10
studied well and performed convincingly (4), that the pianist Jessye Norman 6
and singer should work as a team, being careful not to over- Elisabeth Schwarzkopf 4
shadow the piano (4), and that the emotion and story line that Bryn Terfel 4
is being sung about should be carefully interpreted (8). Again, Elly Ameling 3
the text is emphasized: ‘‘A lieder singer should never perform a Peter Schreier 2
Felicity Lott 2
piece without intimate knowledge of their text - the poetry is the
Janet Baker 2
reason they are there.’’
4 Journal of Voice, Vol. -, No. -, 2015

occurring composers were Schubert (12 participants), Schu- Discussion

mann and Strauss (five participants each), and Brahms (three The responses from the survey highlighted some interesting
participants). opinions about the performance of lieder. The common words
Question 9: ‘‘Select one of the recordings from the above and phrases used indicate that many of the participants placed
question. Why is the performance good?’’ Eleven participants importance on similar aspects. These trends were also reflected
answered this question. A beautiful, flexible, and varied tone in three texts on singing by Elliot, an edited volume by Hall-
from the singer was reported as an important aspect by seven mark and a guide to performance by Miller.23–25
participants. Good diction and language (4) and an interesting The analysis indicated that a ‘‘natural’’ use of vibrato was
and understood interpretation of the text (4) were again high- preferable. Although a specifically ‘‘natural’’ use of vibrato
lighted. An awareness of style and the composer’s intentions for lieder is not explicitly stated in the texts, the results of the
was mentioned specifically by three participants. survey indicate that a supposedly wide or ‘‘operatic’’ use was
Question 10: ‘‘Is there anything you would change about the considered inappropriate. However, as with several comments
performance mentioned above?’’ No participants provided use- made in the self-reported comments and in the literature, a
ful information in this section, and in fact 12 participants stated quantifiable definition of ‘‘natural’’ vibrato (eg, in terms of
that they could not improve the performance, and six partici- vibrato rate, onset, or extent, discussed in the following) was
pants did not respond at all. Participants seemed to be reluctant not forthcoming.
to critique the performers, with one participant noting that they Why are text and diction mentioned so frequently? Can it not
‘‘wouldn’t dare’’ give criticism. Other respondents concluded be assumed that all vocal performances need to have a high
that imperfection is part of what makes performance ‘‘real’’ level of textual comprehension and diction? Possible interpreta-
and that maturity will ‘‘add depth to tone and artistry’’ of the tions could be that in lieder, the voice is more exposed because
performance. of the lower amount of accompaniment in terms of both instru-
Questions 11 to 12: Many of the participants did not answer mentation and potential volume of sound compared with genres
the next three questions (Questions 11, 12, and 13). When such as music theater and opera, therefore creating a need for
asked to name recordings of lieder, they viewed as being perfect diction and a more interesting interpretation. At its
‘‘poor,’’ and to identify what makes it poorly sung, only eight inception, lieder aimed to converge German poetry and music.
participants responded. Participants tended to mention per- The dual focus on text as well as the use of the German lan-
formers rather than specific recordings for this question and guage was innovative at the time. Perhaps, this remains true
did not go further into detail (3). Four participants responded in the performances of today. Furthermore, given that lieder
to Question 12. The disliked features of the poor recordings performers do not have the advantage of props and staging
included a ‘‘tremulous’’ or ‘‘wobbly’’ sound, not being able found in much opera and music theater, clarity of text may
to ‘‘let their egos take a second place,’’ being ‘‘indulgent in receive greater weighting.
the beauty of their own voice,’’ creating a ‘‘boring interpreta- Although agreement was identified in the survey, some of
tion,’’ use of inappropriate singing styles, and a poor use of which was also consistent with the literature, it also became
the German language. apparent that a number of terms were not clearly defined. These
Question 13: ‘‘What would you recommend that this singer include ‘‘good tone,’’ ‘‘subtle vibrato,’’ and ‘‘nonoperatic pro-
change? How would you explain this to the singer?’’ This ques- jection.’’ These findings raised additional questions that were
tion focused on pedagogic implications that could be made to further investigated from an acoustic perspective. Given the
enhance lieder singing. Only six participants answered this prevalence of comparisons reported between lieder and opera,
question, which again reflected the participant’s reluctance to such as the former using less vibrato, and our interest in
comment negatively or constructively on the performances of comparing qualitatively reported characteristics with acousti-
well-known singers. Alternatively, the poor response rate may cally measured variables, we conducted two additional studies.
also be interpreted as a sign of a lack on knowledge on the par- These studies took a more objective approach, measuring
ticipant’s part. Furthermore, placing critique on a professional acoustic properties of singers and comparing lieder and opera
singer may require that the survey respondent hold an equal singing. We limited our investigation to two acoustic parame-
or more advanced knowledge of singing to that singer. A ters of the voice: vibrato and SF.
more thorough and imaginative text interpretation (3) and better
diction (2) were the only recurring comments for this question.
Some of the respondents felt they had answered this question STUDY 2: A COMPARISON OF VIBRATO AND SF
exhaustively in Question 12, and so referred to that question BETWEEN LIEDER AND OPERA
rather than restating their opinions. The aim of study 2 was to determine whether an acoustic anal-
Throughout the survey, some aspects were mentioned repeat- ysis supported the more consistent views raised in the literature
edly. Commonly stated were words such as ‘‘text’’ (67 mentions and the survey. This was done through a comparison of acoustic
across the entire data set) and ‘‘poet’’ or ‘‘poetry’’ (25), characteristics of lieder with opera. Specifically, comparisons
‘‘diction,’’ and other related words such as ‘‘language,’’ were made between SF and three aspects of vibrato: vibrato
‘‘German,’’ and ‘‘pronunciation’’ (50), ‘‘interpretation’’ (20). onset time (delay from onset of a stable note/phoneme to onset
Phrases about the partnership between voice and piano (36) of vibrato), pitch vibrato rate (number of regular pulsations in
and the comparison to opera (24) were mentioned regularly. pitch per second) and vibrato extent (half the maximum to
Lynette Johnson-Read, et al Performing Lieder 5

minimum fundamental frequency fluctuation in vibrato; for single, sustained note was chosen from each excerpt that was
more information see Sundberg19). Based on the literature re- sung on a similar vowel sound, occurred in the same place in
view and qualitative data obtained in study 1, it was hypothe- the overall phrase, and was performed in a comparable register.4
sized that, when compared with opera. These notes were chosen because they were unaccompanied,
H1: SF is lower in lieder excerpts (eg, because of the lack of thus allowing a clear analysis without interference from the
competition from the piano accompaniment, compared with an accompaniment.4 Using a portion of a recitative from a Mozart
orchestra). opera, instead of Wagner for example, provided a conservative
H2: vibrato onset is the same between genres (an interpreta- analysis, in that the Mozart style possesses relatively more sub-
tion based on Miller’s assertion that the tone should not be tle differences in vocal styles compared with lieder than some
started ‘‘straight’’). later, grander operatic forms. This way, we could be confident
H3: vibrato extent is smaller (not as wide) in lieder excerpts. of the veracity of any measurable differences, should any be
In addition, we investigated the vibrato rate between opera identified. A list of the performers and dates of recordings
and lied. However, we were not able to find any literature that used are shown in Table 2.
explicitly predicted if there should be a difference between Procedures. For acoustic measurements, the sound record-
the two genres, and no survey responses specifically mentioned ings were analyzed at each note of interest using sound editing
this characteristic. and analysis software Audacity 1.3.3 Beta (GNU General
Public License software, http://audacity.sourceforge.net/).29
Method Vibrato measurements were made with the spectrogram in Au-
Design. Two designs could be used to investigate the hypoth- dacity, set to a sampling rate of 44100 Hz and a Fast Fourier
eses. One is to recruit singers and ask them to perform in certain Transform (FFT) window size of 2048 samples, using Hann
ways, and an other is to use existing recordings of respected windows. For analysis of SF, measurements were made using
commercially recorded performances. We decided to limit our the frequency spectrum plotter in Audacity, set to a sampling
investigation to existing commercial sound recordings of well- rate of 44100 Hz and a fairly course window size of 128 sam-
known performers to avoid the problems of the former, being ples (344 Hz) for the easy identification of formants, rather
that either students would have to be employed, possibly than the details of the partials.
reducing the control we had over different performance condi- Vibrato parameters were calculated by inspection of the spec-
tions (lied vs opera), or the prohibitive cost and time of employ- trogram. Vibrato rate was measured by identifying each full
ing highly respected performers. Furthermore, the difficulties of vibrato cycle, which consisted of a consecutive peak and trough,
instructing any singer to perform a lied and an opera extract may and the total number of these cycles was counted and then
reveal the design problem of demand characteristics, where the divided by the duration of those cycles in seconds. This produced
researcher cannot be sure that the singing participant is not ad- a vibrato rate in hertz. Vibrato extent was estimated by reading
justing their performance as an artifact of the experimental con- the peak and trough frequencies during a region of sustained
dition, rather than as an effect of the different genres. Although vibrato. The peak frequency was divided by the trough fre-
recruiting singers may be a tempting and interesting avenue for quency, halved, and then converted into cents (a unit related to
further study, we decided to use commercial recordings instead. pitch that divides an equal temperament semitone into a further
However, even conducting acoustic analysis using commercial 100 units), as is the convention for vibrato extent22 (see
recordings presented several design challenges. Among the Figure 2 for an example of the readings taken to make these
most important was the acoustic separation of vocal information calculations).
from piano/orchestral accompaniment, which is generally hard The relative level of the SF (henceforth LSF) was quantified
to do with commercial sound recordings. With this in mind, by subtracting the sound level of the first formant (henceforth,
we opted to choose sound recordings from lieder and opera formants are denoted as Fn, where n ¼ formant number, and
where at least one note could be identified that had minimal their levels are measured in units of dB) from the sound level
interference from the audio signal of the accompaniment and of the SF (also in dB). Figure 3 illustrates how the calculation
we performed acoustic analyses on these. was read from the spectrum. This provided information about
the relative strength of the SF, based on methods used by Ross-
Stimuli. Altogether, 38 single note excerpts were examined. ing et al30 and Omori et al31 and removed the problem of using
Specific choices, such as the voice type and composer, were SF readings that were misleading because of overall sound
guided by the results of the survey (Table 1). Baritones were recording level. We delve into more detail in the measurement
chosen because a high number of performers with that voice of LSF in Study 3.
type were identified throughout the survey, and the literature
associates SF with lower voice types.28 A single unaccompa-
nied note from each of Franz Schubert’s songs Wasserflut Results
and Der Erlk€onig were used as the two lieder excerpts Results of the acoustic measurements are summarized in
(Figure 1A and B). Two single notes from the extended recita- Table 3. The overall strongest LSF was in the Tahu Rhodes
tive before the Count’s aria ‘‘Vedr"o mentr’io sospiro’’ in W. A. A3 operatic note (3 dB higher than F1; Figure 3), and the weak-
Mozart’s The Marriage of Figaro provided two excerpts for est LSF was found in Terfel’s Der Erlk€onig (29 dB below F1).
comparison from the operatic repertoire (Figure 1C and D). A On average, the opera excerpts had a stronger LSF (6 dB below
6 Journal of Voice, Vol. -, No. -, 2015

FIGURE 1. A. Extract from Schubert’s Wasserflut, depicting one of the lieder examples used in study 2. B. Extract from Schubert’s Der Erlk€onig,
depicting one of the lieder examples used in study 2. C. Extract from Mozart’s Le Nozze di Figaro, depicting one of the opera examples used in study 2.
D. Extract from Mozart’s Le Nozze di Figaro, depicting one of the opera examples used in study 2.

F1) than the lieder excerpts (12 dB below F1). On average, Mean vibrato onset time was much shorter for operatic ex-
vibrato extent was wider for opera excerpts (148 cents, about cerpts compared with lieder excerpts (63 milliseconds and
one and a half semitones) than for lieder excerpts (106 cents, 276 milliseconds, respectively), but it should be noted that
about one semitone). note durations for the Wasserflut excerpt were considerably
Lynette Johnson-Read, et al Performing Lieder 7

TABLE 2.
Performers and Year of Recording for the Excerpts Used for the Acoustic Analysis in Study 2
Le Nozze di Figaro
Wasserflut € nig
Der Erlko (M1 and M2)
Dietrich Fischer-Dieskau (1988) Dietrich Fischer-Dieskau (1988) Dietrich Fischer-Dieskau (1968)
Dietrich Henschel (2006) Bryn Terfel (2007) Bryn Terfel (2006)
Hermann Prey (2008) Hermann Prey (2008) Teddy Tahu Rhodes (2003)
Gerhard Hu € sch (1999) Gerhard Hu € sch (2001) Thomas Allen (1982)
Matthias Goerne (2005) Matthias Goerne (1997) Thomas Hampson (1991)
Patrick Mason (1995) Ernst Buscagne (2002) Jonathan Lemalu (2005)
Jon Vickers (1992) Marko Rothmu € ller (2008) Jose# Van Dam (2005)
Thomas Hampson (2005) Wolfgang Holzmair (1993) Bo Skovhus (2007)
Carl-Heinz Mu € ller (1988) — Alfred Poell (1955)
Olaf Ba€r (2007) — William Stone (2009)

longer. Der Erlk€onig had a more comparable duration with the The results of these acoustic vibrato onset measurements
Mozart excerpts (average duration 694 milliseconds), and the indicate that, in comparison with opera, vibrato in lieder begins
corresponding vibrato onset time was 135 milliseconds. On later within a note. In this study, Henschel’s performance of the
examining the proportional time of entry of the onset with note of interest in Wasserflut had the most delayed vibrato onset
respect to the duration, this gives a ratio of 135/ of 1.85 seconds for a note of duration 2.85 seconds. The average
694 ¼ 0.195. For the opera excerpts, this ratio is 63/ delay of vibrato in lieder technique suggests that vibrato may be
533 ¼ 0.118 (Table 3), meaning that on average, lieder notes used as a coloration rather than as an immediate feature of vocal
started vibrato proportionally later (with respect to note dura- tone. This finding therefore questions the negative connotation
tion) than opera notes. Vibrato rate was similar across the two in the use of delayed vibrato onset. As discussed in the introduc-
genres (6.16 and 6.20 Hz for lieder and opera notes respec- tion, a delay in vibrato onset has been argued to be indicative of
tively). LSF and vibrato extent, onset time, and rate were sub- a faulty technique, for example where Miller25 states that a lie-
mitted to Mann-Whitney U tests. LSF, vibrato extent, and der singer should not let the vocal tone start straight and then let
vibrato onset time were all significantly different between it ‘‘wiggle with vibrato’’ (p. 16). It must be noted, however, that
the two genres at P ¼ 0.01, whereas vibrato rate was not Miller may have been referring to the avoidance of extreme use
(Table 4). of vibrato per se, with which our data concur. The contribution
of the present analysis allows for a more specific acoustic
Discussion description of lieder performance. Furthermore, delay in
Using acoustic analysis techniques, three hypotheses were vibrato onset may be an acoustic representation that describes
tested in study 2. Two were supported, one rejected: lieder the idea of pure tone reported in study 1.
excerpts were performed with a weaker LSF (H1) and The estimated vibrato extent showed that a narrower vibrato
smaller vibrato extent (H3) than the opera excerpts. Howev- of around 100 cents was used in the lieder excerpts, compared
er, there was a difference in vibrato onset time, contrary to with an average of 150 cents in opera. These values are consis-
H2, with a longer delay in lieder than in opera. Additionally, tent with those of Prame.22 This is also consistent with reports
no difference was observed in vibrato rate between the made in both the survey and literature. Examination of the two
genres. sampled performers who had recordings in both genres,

FIGURE 2. Spectrogram analysis from study 2 using Audacity to identify vibrato parameters.
8 Journal of Voice, Vol. -, No. -, 2015

FIGURE 3. Spectral analysis in study 2 using Audacity to measure LSF.

Fischer-Dieskau and Terfel, demonstrated a distinction in ment, a primary focus on text and diction in lieder compared
vibrato onset delay across the two genres. For lieder excerpts, with opera could be another explanation for why SF is lower.
the delays were 100, 60, and 400 milliseconds, versus 50, 30,
and 20 milliseconds for opera, whereas other parameters
measured varied less systematically (Table 3). STUDY 3: COMPARISON OF SINGERS WHO
SF was identified in both the opera and lieder recordings. Its PERFORM IN BOTH GENRES
use on a held note at the end of a phrase in Wasserflut is apparent One of the criticisms of the design of study 2 is that not all
in a majority of the recordings. However, the statistical analyses singers performed in each genre. It can be argued that the results
suggest that there is an acoustically significant difference be- were confounded by variations across performer rather than
tween the two genres, with lieder singers being less reliant on across genre. Furthermore, no explicit attempt was made to
the strength of that formant. Again comparing the performers control for the duration or pitch of notes because of the obvious
who were sampled in both genres, Fischer-Dieskau and Terfel, difficulties in doing so, particularly with a small sample. To
these excerpts exhibited a stronger LSF in the opera excerpts address these limitations, a third study examined only singers
(#7.8, #3.8, #7.4, and #8.3 dB) than in the lieder excerpts who had recordings available for both opera and lied. The num-
(#12.8, #5.9, and #29.4 dB). It is apparent from these data ber of excerpts used for analysis was also increased. Compari-
that professional, commercially recording singers have consid- sons were made to further ascertain whether characteristics in
erable flexibility in adapting to the different genres. The results SF and vibrato differed because of the genre performed.
are also consistent with Sundberg and Romedahl’s3 observation The three hypotheses proposed in study 2 were also tested in
that a lack of SF increased text intelligibility. Furthermore, this study. Additionally, this study included a fourth hypothesis
beyond not needing SF to project over the lighter accompani- that there would be no difference in mean vibrato rate between
Lynette Johnson-Read, et al Performing Lieder 9

TABLE 3.
One-Note Analyses of Singer’s Formant Strength, and Vibrato Extent, Onset and Rate, for Lieder Versus Opera
SF Vibrato Note Vibrato Onset Vibrato
Source Performer Year Strength (dB) Extent (cents) Duration (ms) Delay (ms) Rate (Hz)
€ nig
Der Erlko Buscagne 2002 #3.1 59 600 130 6.38
Fischer-Dieskau 1988 #12.8 165 700 60 6.25
Goerne 1997 #8.9 113 800 150 5.88
Holzmair 1993 #15 130 350 50 6.67
Husch 2001 #18.8 142 870 130 6.76
Prey 2008 #8.6 75 330 20 6.45
Rothmuller 1944 #8.4 41 750 140 6.56
Terfel 2007 #29.4 132 1150 400 6.67
Wasserflut Baer 2007 #9.4 86 3150 100 6.39
Fischer-Dieskau 1988 #5.9 95 3600 100 5.85
Goerne 2005 #11.8 104 3000 350 6.04
Hampson 2005 #6.5 116 2300 50 5.71
Henschel 2006 #9.2 131 2850 1850 5.50
Husch 1999 #13 77 2800 450 6.38
Mason 1995 #11 93 3500 150 6.00
Muller 1988 #6.5 92 3080 80 5.70
Prey 2008 #8.3 158 3800 150 5.62
Vickers 1992 #22.2 106 4300 600 6.10
Mean (SD) of all Lieder #11.6 (6.42) 106 (33) 2107 (1377) 276 (424) 6.16 (0.394)
M1 (D4) Allen 1982 #9.1 119 650 50 6.67
Fischer-Dieskau 1968 #7.8 151 550 50 6.22
Hampson 1991 #8.3 98 610 40 6.45
Lemalu 2005 #5.2 185 700 100 6.67
Poell 1955 #4.9 196 350 100 6.50
Skovhus 2007 #5.1 146 260 50 5.98
Stone 2009 #12 129 900 150 5.93
Tahu Rhodes 2003 #2 106 470 40 5.97
Terfel 2006 #7.4 177 220 30 6.71
Van Dam 2005 #4.1 178 250 80 5.88
M2 (A3) Allen 1982 #8.4 146 660 30 5.56
Fischer-Dieskau 1968 #3.8 175 750 50 6.43
Hampson 1991 #3.7 124 400 10 6.30
Lemalu 2005 #3 193 750 50 6.11
Poell 1955 #0.3 172 400 50 5.71
Skovhus 2007 #7.7 64 220 20 6.28
Stone 2009 #16.1 163 940 30 6.35
Tahu Rhodes 2003 3.3 170 650 150 6.30
Terfel 2006 #8.3 100 380 20 5.56
Van Dam 2005 #3.8 172 550 160 6.41
Means (SD) of all Opera #5.885 (4.212) 148 (36.5) 533 (220) 63 (45.5) 6.20 (0.351)

the two genres (H4). H4 aims to reproduce the results of study 2, LSF in relation to the overall sound level of a sung pitch, we
in regards to vibrato rate. Furthermore, because the published decided to explore three methods, which are presented in the
literature has not reached a consensus on how to best measure Procedures of the Method section.

TABLE 4.
Mann-Whitney U Tests for SF Strength, Vibrato Extent, and Vibrato Onset Time With Genre (Lieder or Opera) as the
Independent Variable
SF Strength Vibrato Extent Time Until Vibrato Onset Vibrato Rate
Mann-Whitney U 68.5 69 70.5 170
Z #3.261 #3.245 #3.223 #0.293
Asymp. Sig. (2-tailed) 0.001 0.001 0.001 0.77
Abbreviation: Asymp. Sig., asymptotic significance.
10 Journal of Voice, Vol. -, No. -, 2015

Method with a Welch frequency spectrum using a Hann window. An

Stimuli. One hundred excerpts, all performed by baritones, FFT window size of 1024 samples was used for F0, and 512
were selected for analysis of SF (Table 5). Inclusion criteria samples for F1–F5. Stimuli were sampled at a rate of
were as follows: 22050 Hz, and with a sample duration of 200 milliseconds.
The spectra used a 20% overlap between segments. Frequency
1. The performers had to be highly acclaimed, international and intensity levels for F0 and F1–F5 were determined using the
professionals in both lied and opera. findpeaks function. The F0, formant frequency, and time read-
2. The recordings were available. ings were transferred onto a spreadsheet before statistical ana-
3. Sung pitches could be identified that had minimal or no lyses. All statistical analyses were performed using SPSS
simultaneous accompaniment (eg, orchestral or piano). statistics package version 22.33
Intensity levels for F0, and F1–F5 were recorded in decibels;
All selected excerpts consisted of a single note with a mini- however, F3, F4, and F5 were not always clearly identifiable. To
mum duration of 500 milliseconds. The sung pitches in the SF aid with identification of ambiguous excerpts, F3, F4, and F5
analysis ranged from Ab2 F0(!103 Hz) to G4 (F0!392 Hz). A were assumed to lie in the region of the frequency spectrum
second set of 90 excerpts from the same six performers were where the SF is expected, from 2200 Hz to 4500 Hz.2 F1 was
selected for vibrato analysis (Table 6). Of these 90 excerpts, assumed to lie in the region of vowel definition, up to 900 Hz
86 were taken from the initial batch of 100 SF excerpts, with in the frequency spectrum.34 Given the variety of options for
14 SF excerpts deemed unsuitable for vibrato analysis due to quantifying SF, three methods were applied here, which we
interference from accompaniment. The additional four excerpts refer to as A, B, and C.
added to the vibrato set were taken from the same contingent of
performers and pieces. The inclusion criteria for vibrato ex- $ Method A compared the level of the putative (visually
cerpts were identical to that of the SF stimulus set. The sung obvious) peak in the spectral envelope in the region of
pitches in the vibrato analysis ranged from F# (F0!174 Hz) vowel definition (henceforth ‘vowel putative’) with the
to G#4 (F0!415 Hz). level of the putative peak in the region in which SF
Procedures. Excerpts meeting the selection criteria were ex- (henceforth ‘SF putative’) is visually expected. The differ-
tracted and saved as separate audio files. Each excerpt consisted ence in level between these two points was recorded as the
of a single vowel, plus consonants, performed on a single pitch, LSF in units of decibels, as summarized in Figure 4.
as per study 2. Excerpts from the SF stimulus set were analyzed $ Method B is identical to the method used in study 2. This
at each note of interest using the frequency spectrum analyzer method compared F1 instead of vowel putative with the
routine in MATLAB R2013b.32 Spectral analysis was performed SF putative as summarized in Figure 4.

TABLE 5.
Performers and Works From Which Excerpts Were Selected for SF Analysis in Study 3 (Lied Compared With Opera)
Number
Performer Piece Genre of Excerpts
Bryn Terfel Der Erlko€ nig Lied 3
Bryn Terfel Schwanengesang Lied 3
Bryn Terfel Der Fliegende Holla€ nder Opera 9
Bryn Terfel Don Giovanni Opera 2
Bryn Terfel Le Nozze di Figaro Opera 5
Dietrich Fischer-Dieskau Schwanengesang Lied 2
Dietrich Fischer-Dieskau Winterreise Lied 7
Dietrich Fischer-Dieskau Der Fliegende Holla€ nder Opera 8
Dietrich Fischer-Dieskau Le Nozze di Figaro Opera 4
Hermann Prey Morgen Opera 5
Hermann Prey Winterreise Lied 6
Hermann Prey Le Nozze di Figaro Opera 5
# Van Dam
Jose Schwanengesang Lied 2
# Van Dam
Jose Winterreise Lied 4
# Van Dam
Jose Der Fliegende Holla€ nder Opera 10
Matthias Goerne Schwanengesang Lied 2
Matthias Goerne Winterreise Lied 4
Matthias Goerne Le Nozze di Figaro Opera 5
Matthias Goerne Tannha € user Opera 2
Thomas Allen Winterreise Lied 5
Thomas Allen Don Giovanni Opera 2
Thomas Allen Le Nozze di Figaro Opera 5
Lynette Johnson-Read, et al Performing Lieder 11

TABLE 6.
Performers and Works From Which Excerpts Were Selected for Vibrato Analysis in Study 3 (Lied Compared With Opera)
Number
Performer Piece Genre of Excerpts
Bryn Terfel Der Erlko€ nig Lied 2
Bryn Terfel Schwanengesang Lied 5
Bryn Terfel Der Fliegende Holla€ nder Opera 9
Bryn Terfel Le Nozze di Figaro Opera 1
Dietrich Fischer-Dieskau Schwanengesang Lied 3
Dietrich Fischer-Dieskau Winterreise Lied 7
Dietrich Fischer-Dieskau Der Fliegende Holla€ nder Opera 7
Dietrich Fischer-Dieskau Le Nozze di Figaro Opera 3
Hermann Prey Morgen Lied 5
Hermann Prey Winterreise Lied 6
Hermann Prey Le Nozze di Figaro Opera 4
# Van Dam
Jose Winterreise Lied 4
# Van Dam
Jose Der Fliegende Holla€ nder Opera 10
Matthias Goerne Schwanengesang Lied 1
Matthias Goerne Winterreise Lied 6
Matthias Goerne Le Nozze di Figaro Opera 5
Matthias Goerne Tannha € user Opera 2
Thomas Allen Winterreise Lied 5
Thomas Allen Le Nozze di Figaro Opera 5

$ Method C compared the level of F1 with the level of F3. peaks are present in either of these spectral areas. In this
The difference in level between these two points was re- circumstance, Schutte and Miller replaced the putative peak
corded as the LSF in units of decibels as summarized in level with the level half way between the two highest peaks.
Figure 4. Schutte and Miller do not outline any further criteria for identi-
fying exactly how prominent a secondary peak must be or
Method A is similar to one used by Schutte and Miller.35 discuss steps for encountering a larger number of competing
Schutte and Miller used a vowel putative and SF putative; how- peaks. When compared with method A, we feel this is a limited
ever, their method differs to method A when multiple prominent method for measuring LSF, as both the spectral areas used for

FIGURE 4. An example of methods A, B and C, as used to calculate LSF in study 3. This example depicts an excerpt from Der Erlk€onig, as per-
formed by Bryn Terfel.
12 Journal of Voice, Vol. -, No. -, 2015

the vowel putative and SF putative will often be comprised of in this study. This lower sample size often yielded two broad
several peaks. We argue that the use of a single measurement peaks in the spectral envelope, labeled as F1 and SF, as opposed
from the most prominent peak is a more reliable point for com- to a collection of partials across the spectrum. The deliberately
parison among a large number of stimuli than an average of low frequency resolution of this method avoids the confusion of
multiple peaks. partials and formants and negated the need for putative peaks.
In a study such as this, in which varying vowel sounds are However, the low frequency resolution can also be seen as pre-
compared, we propose that methods without a vowel putative senting lower precision.
may suffer from inconsistencies. Methods B and C fall into Excerpts on the vibrato list were analyzed at each note of in-
this category. Specifically, any change in the positioning of terest using the MIRtoolbox 1.3.3 in MATLAB R2013b.32 Pitch
the lips, jaw, tongue, velum, and larynx can affect the amplitude vibrato measurements were determined with the mirspectrum
and frequency of F1.34 Furthermore, although F1 commonly ex- spectrogram using an FFT window length of 1024 samples
hibits the highest intensity level in the area of the spectrum near and a Hamming window. The spectrogram frame ranged from
F0, which is often the basis for its use as a reference point, this is 0 to 3500 Hz. Peaks and troughs were determined with the gin-
not always the case. It is possible that a certain vowel sound put function. The pitch and time readings were transferred onto
with a lower F1 level could be measured at a lower intensity a spreadsheet prior to statistical analyses. All statistical ana-
than the reading F0 or another prominent peak of the same lyses were performed using SPSS statistics package
vowel would give. In this situation, we argue a vowel putative version 22.33 Vibrato calculations were made following the
has greater flexibility. We do, however, acknowledge that in a same methods as study 2 with an example shown in Figure 5.
study comparing examples of matched vowel sounds, in which
F1 could be expected to be similar across examples, vowel mea-
surement based solely on F1 could prove more reliable. Results
Method C may also produce an inaccurate representation of Results pertaining to hypotheses tested. SF was
the overall level of the SF due to the potential distance from the observed in all 55 opera excerpts and in 40 of the 45 lieder ex-
actual, though not graphically visible SF putative. This inaccu- cerpts. All three methods yielded a stronger mean LSF for oper-
racy arises from the formation of the SF, which is sometimes atic excerpts compared with lieder excerpts. The results of
proposed to be a clustering of F3, F4, and F5.2 As an example, method A are summarized in Figure 6.
the SF clustering in Figure 4 shows F3 at a lower level than F4 In regards to method A, the results of five of the six singers
and could therefore be an inaccurate estimate of the level of SF show a trend in which operatic performances feature a stronger
intensity, which by simple visual extrapolation appears to be mean within-singer SF than the lieder performances. These ob-
closer to the intensity of F4. servations support the view that the SF is present in lieder per-
Additionally, the frequency spectra used in study 2 featured a formance, albeit in a slightly weaker form. Goerne featured a
significantly lower window size (128 samples) than those used higher LSF in lieder excerpts; however, these results may be

FIGURE 5. Spectrogram analysis used to identify vibrato parameters in study 3. This example depicts an excerpt from Schubert’s Die Wetter-
fahne, as performed by Thomas Allen.
Lynette Johnson-Read, et al Performing Lieder 13

FIGURE 6. Summary of LSF results in study 3, according to method A. Results are displayed by performer and genre. Error bars ¼ ±1
standard error.

because of the low number of excerpts used in each genre for ference was statistically significant according to a t test
this singer. (t(88) ¼ #2.153, p ¼ 0.007, but note that the assumptions of
To examine whether any of the SF and vibrato parameters normality for these temporal data are quite likely violated and
were different between lieder and opera performance, a series should be treated with caution). On examining the proportional
of t tests were conducted. Because of the low number of within time of entry of vibrato onset with respect to note duration
performer excerpts, a meaningful within-singer paired compar- (vibrato onset time/note duration time), the ratio was 0.11 for
ison was considered inappropriate. Instead, these observations lieder and 0.07 for opera. This means that, on average, lieder ex-
were further investigated through inferential statistical methods cerpts began vibrato proportionally later (with respect to note
on a pooled genre-by-genre basis. These results are summarized duration) than opera excerpts. That is, typically 11% of the
in Table 7. note has elapsed before vibrato commences in a lied note,
Using LSF method A, M ¼ #38.5 dB, standard deviation whereas only 7% of the note has elapsed in an opera note. These
[SD] ¼ 6.32, n ¼ 55 for opera excerpts which is higher than results correspond with the findings of study 2, in which onset
M ¼ #41.8 dB, SD ¼ 7.0, n ¼ 45 for lieder excerpts. The dif- time was also statistically different between the genres.
ference in LSF between the two genres was statistically signif- Operatic excerpts featured a slightly greater vibrato rate
icant according to an independent samples t test (t(98) ¼ 3.209, (M ¼ 6.8 Hz, SD ¼ 1.9, n ¼ 46) than lieder excerpts
p ¼ 0.002). These results are summarized in Figure 7. (M ¼ 6.4 Hz, SD ¼ 1.1, n ¼ 44). In similar fashion to the results
Method B yielded M ¼ #32.7 dB, SD ¼ 6.7, n ¼ 55 for opera of study 2, this study was not able to identify a difference in
excerpts and M ¼ #36.5 dB, SD ¼ 7.2, n ¼ 45 for lieder vibrato rate between the two genres with statistical significance
excerpts. The difference in LSF between the two genres was according to an independent samples t test (t(88) ¼ 1.205,
statistically significant according to an independent samples t p ¼ 0.231). Additionally, in concert with study 2, the average
test (t(98) ¼ 3.26, p ¼ 0.002). Method C yielded vibrato extent was wider in operatic examples (M ¼ 127 cents,
M ¼ #37.5 dB, SD ¼ 6.1, n ¼ 55 for all opera excerpts and SD ¼ 56, n ¼ 46) than in lieder excerpts (M ¼ 118 cents,
M ¼ #40.3 dB, SD ¼ 8.3, n ¼ 45 for all lieder excerpts. The SD ¼ 49, n ¼ 44). No significant difference was found in
difference in LSF between the two genres was statistically sig- vibrato extent between the two genres according to an indepen-
nificant according to an independent samples t test dent samples t test (t(88) ¼ 0.867, p ¼ 0.256). The difference in
(t(98) ¼ 2.302, p ¼ 0.023). vibrato extent between the two genres was slightly less than in
The results of method A showed the lieder excerpts to have a study 2, which featured an average extent of 148 cents in opera
mean LSF 3.3 dB weaker than the operatic excerpts. The results and 106 cents in lieder. The lower values could be an artifact of
of method B showed the lieder excerpts to have a mean LSF the slight differences in signal processing and analysis software
3.8 dB weaker than the operatic excerpts, and method C showed used across the two studies.
the lieder excerpts to have a mean LSF 2.8 dB weaker than the
operatic excerpts. In comparison, the results of study 2 showed Comparison of three methods for calculating level of
the lieder excerpts to have a mean LSF 5.7 dB weaker than the the SF. The three methods used to calculate LSF produced
operatic excerpts. Overall, the mean relative SF intensity is fairly consistent results with each other, and they were consistent
significantly lower in lieder compared with opera by approxi- with the calculations in study 2 although small discrepancies
mately 3 to 5 dB regardless of the method used to perform could be explained by the different software (Audacity in study
the calculation. 2, MATLAB in study 3) and software settings used across the
Overall, vibrato onset time was shorter for the operatic ex- two studies. Methods A and B produced comparable results, in
cerpts (M ¼ 80 ms, SD ¼ 110, n ¼ 46) when compared with which the mean LSF of each singer was consistently weaker in
the lieder excerpts (M ¼ 1605 ms, SD ¼ 180, n ¼ 44). This dif- lieder by a similar difference. Method C produced LSF readings
14 Journal of Voice, Vol. -, No. -, 2015

that were slightly closer between the two genres. Additionally,

1450 (991.31) 117.77 (48.42) 6.38 (1.05)

127.39 (55.23) 6.77 (1.86)

when examining LSF for individual singers, methods A and B pro-
Vibrato
Rate

5.43
6.36
7.09

6.75
6.12

6.97
6.33
6.21
6.57
6.96
(Hz)

duced relatively higher values of LSF in opera for five of the six
5.8

7.8
singers, whereas method C produced results in which only three
singers exhibited a stronger LSF in opera and the remaining three
a stronger LSF in lieder. The results of Goerne’s performances,
Vibrato

which had a stronger LSF in lieder regardless of the method

(cents)
165.44
124.07
105.91

103.66
110.79

164.86
114.46

138.47
130.02
84.59

84.43
Extent

86.8
used, can be attributed to a personal technique choice by this
performer. However, we decided to perform a visual spectral
investigation on the results of Terfel and Prey, who showed a
stronger LSF in opera for methods A and B and the opposite for
Mean Note

method C. That is, method C produced the most conservative

Duration

1241 (754)
2240
1540

1880
1430
1450

1207
1783
1271
1056

1152
820

822
(ms)

readings (identifying fewer differences), but with some inconsis-

tencies for the Terfel and Prey excerpts.
Although all three methods showed a statistically significant
difference in LSF between the two genres for the pooled data,
Mean Vibrato

an investigation of the spectra revealed that method C may be

Mean (SD) of all Lieder #41.8 (6.93) #36.5 (7.24) #40.3 (8.34) 156.25 (176.92) 0.113 (0.106)
Onset Time

0.07 (0.08)

less accurate than either of the other two methods that used pu-
0.116
0.101
0.132
0.102
0.135
0.065

0.019

0.151
0.073
0.105
0.029
Ratio

0.08

tative peaks in the calculation. This stems from the use of only
F3 to measure the entire SF intensity, as discussed in the
Method section. Of the 22 excerpts performed by Terfel, the
level of F3 appeared to be a relatively inaccurate representation
83.9 (107.44)
Mean Vibrato

Time (ms)

of the overall SF in 10 examples (44% of the excerpts). Simi-

133.76
111.35
199.28
217.04

119.69
183.65

107.71
84.62

21.89

74.75

41.54
Onset

larly, with the 16 excerpts performed by Prey, spectral investi-

224.7

gation revealed F3 to be a relatively inaccurate indicator of the

overall SF in seven examples (45% of the excerpts). Two of
these examples are displayed in Figure 8.
A second investigation of the spectra was also conducted to
Mean (SD) of all Opera #38.5 (6.32) #32.7 (6.68) #37.5 (6.14)
Mean LSF
(Method

ascertain the accuracy of method B compared with method A.

C, dB)
#35
#40
#34
#44
#42
#47

#35
#35
#37
#34
#46
#38

Although F1 is often the highest intensity formant36 in the

vowel region, it is not unusual for F0 or another peak to domi-
nate. In 40 of the 100 excerpts used in method B, F1 was signif-
icantly lower level than F0. These 40 excerpts were spread
Mean LSF
(Method
B, dB)

relatively evenly across the two genres. They accounted for

#33
#37
#31
#42
#37
#39

#30
#33
#26
#32
#40
#35

14 out of the 38 lieder excerpts (37%), and 26 out of 62 opera

excerpts (42%). An example of this is displayed in Figure 9.
With this in mind, we argue that method A corresponds most
closely with the visually apparent location of the SF of the three
Mean LSF
(Method
A, dB)

methods. This largely stems from the flexibility the method is

#37
#41
#41
#47
#42
#43

#36
#38
#32
#38
#48
#39

afforded from having two putative peaks and is therefore able

to produce plausible results regardless of whether F1 is the
most prominent vowel peak or not. Method C, then, may also
Summary of Acoustic Analysis for Study 3

Dietrich Fischer-Dieskau

Dietrich Fischer-Dieskau

be viewed as a more conservative approach, but further research

will be needed to determine the validity and reliability of the
different methods in more detail.
Matthias Goerne

Matthias Goerne
Vocalist

# Van Dam

# Van Dam
Hermann Prey

Hermann Prey
Thomas Allen

Thomas Allen

Discussion
Lieder excerpts Bryn Terfel

Opera excerpts Bryn Terfel

excerpts

excerpts

Study 3 investigated three hypotheses that were tested in study 2

and that were based on the literature review and qualitative find-
Jose

Jose

ings of study 1. A fourth hypothesis was included in an attempt

to reproduce the vibrato rate results of study 2. In all four hy-
potheses, the findings of study 2 were replicated. As with study
2, a stronger SF was observed in opera excerpts (H1), and a
TABLE 7.

shorter mean onset time was observed in opera examples

(rejecting H2). A slightly wider mean vibrato extent for opera
examples was found, in accordance with H3. No difference
was observed in mean vibrato rate between the two genres (H4).
Lynette Johnson-Read, et al Performing Lieder 15

FIGURE 7. Summary of LSF results in study 3, displayed by genre and method. Error bar ¼ ±1 standard error.

With the replication of the findings, and consistency across It is through this spectral ‘‘gap’’ that a combination of SF and
studies, we need to explain why lied singers do not use SF to vibrato presents an alternative and contrasting path for the
the same extent as opera singers, even if the same performer singer to allow identification of their vocal line. The findings
is observed. Explanations from study 1 suggest that opera from study 1 also suggest that a lesser use of the SF found in
singers include SF in lieder to retain the specific tonal qualities lieder makes the voice more speech-like, allowing greater focus
of the technique; however, the smaller scale of accompaniment on the communication of the words in a more natural language
in lieder requires less voice projection and therefore a lesser de- way that is required in opera.
gree of SF. Importantly, when a singer is competing with an or- In both studies 2 and 3, mean vibrato onset time was found to
chestra, they require additional resources apart from sheer be significantly shorter in opera excerpts. However, not all
intensity as the required intensity could not be achieved against singers conformed to this finding. Prey produced a longer
large orchestral forces accompanying at high volume. The mean vibrato onset time in opera (180 milliseconds) compared
singer is more likely to project over the orchestra by applying with lieder (110 milliseconds). This may be explained by
a stronger SF as well as a near continuous vibrato and wider the considerably shorter mean duration of Prey’s lieder ex-
vibrato extent. Analyses have revealed that SF occupies a spec- cerpts (820 milliseconds) compared with opera excerpts
tral region in which orchestral intensity is relatively weak.36,37 (1708 milliseconds). However, examining the temporal ratio

FIGURE 8. Two spectral examples in which the level of F3 is at a significantly lower level than the rest of the SF (measured here with an SF
putative). The example on the left is an excerpt from Don Giovanni, as performed by Bryn Terfel. The example on the right is an excerpt from
Le Nozze di Figaro, as performed by Hermann Prey.
16 Journal of Voice, Vol. -, No. -, 2015

FIGURE 9. A spectral example in which F1 is not the dominant peak in the vowel region. This example depicts an excerpt from Der Fliegende
H€ollander, performed by Jos#e Van Dam. Measurements are shown for F0, F1, and F2 (left to right).

of vibrato onset time with respect to the overall duration of the of creating a varied channel of pitch delivery, allowing audience
note, hence adjusting for possible biases caused by comparing detection of that voice through auditory stream analysis.39
notes of different duration, we find a rather similar ratio of 15% Expressly, vibrato and SF address practical performance issues.
vs 13% for opera versus lied respectively (Table 7). Similarly, Aesthetic issues may emerge as a result of the success or other-
Fischer-Dieskau also demonstrated a longer mean onset time wise of solving the practical problems. That is, if ‘‘more’’
in opera (130 milliseconds) compared with lieder (120 millisec- vibrato allows for clearer communication in opera, the use of
onds) although the difference here is much less pronounced. vibrato may then become the aesthetic preference in opera as
The vibrato onset ratio is again very similar (8% for opera well.
and 10% for lied). This too can be attributed to note durations, As with study 2, mean vibrato rate was found to be similar
in which the lieder excerpts were on average slightly shorter between the two genres. The slight difference in rate between
(1540 milliseconds) than opera (1780 milliseconds). With this study 2 and study 3 is comparable with those found by Prame21
in mind, the data strongly suggest a connection between the per- although that study did not use examples of opera or lied.
formance of lied and a delayed vibrato onset, in relation to the
length of the overall note duration.
Mean vibrato extent was shown to be slightly wider in the op- GENERAL DISCUSSION AND CONCLUSION
era examples. This result is consistent with other studies, such This research demonstrated that several of the assumptions
as one8 in which vibrato extent was found to be greater in opera made by teachers and performers about the techniques that
(!98 cents) than Broadway musical vocal styles (!78 cents). are appropriate for the performance of lieder can be investi-
The difference in vibrato extent also supports the findings of gated through acoustic measurement. An important contribu-
study 2 although the results in this study are not as pronounced. tion made in the present research is that it has enabled, for
The overall wider extent and earlier onset of vibrato relative to the first time, a systematic scientific approach to examining
lied singing found in this article suggests another strategy some of the features associated with good lieder technique
singers use to compete with the larger forces of their accompa- and facilitates clearer definitions that may be applied by singers
niment, in addition to a strong SF. Orchestral instruments usu- and teachers of the genre. Although previous qualitative reports
ally have vibratos with a rather smaller amplitude than that of of singing style provide generic explanations with poorly
the voice.38 Furthermore, an tutti orchestra almost never coor- defined terms, such as ‘‘natural vibrato,’’ we are able to propose
dinates vibrato. Consequently, a singer with a large amplitude here some of the ways that these descriptions may manifest
vibrato, maintained for a relatively long fraction of the note, themselves in the physical, acoustic properties of a voice aim-
gives the audience a better chance to identify the singer. This ing to produce an appropriate and perhaps even good perfor-
is both through contrast with the relatively stable tones of the mance of lieder, eg, in terms of vibrato ‘‘width’’ (in terms of
orchestra, and possibly also through the perceptual advantage pitch) and amount of delay before the onset of vibrato for a
Lynette Johnson-Read, et al Performing Lieder 17

particular pitch, vowel, or phoneme. We suggest that when a that one should not assume one method of investigation is sov-
vocal pedagogue refers to purity of tone, from an acoustic ereign. Acoustic measurement and the accounts of teachers and
perspective, this corresponds to a delayed vibrato onset time scholars are all important kinds of data that can be considered in
with a narrower vibrato width compared with vocal styles that helping to build a rounder picture of what goes on in the process
use a larger, louder combined accompanying force. Relatively of producing lieder, and for that matter, good lieder.
lower relative intensity of the SF may also contribute to the ped- We recommend further research on this topic, preferably
agogues concept of purity of tone. with more comparable matching in regards to note duration,
By marrying the quantitative and qualitative data, we are able pitch, and the nature of the phonemes sung. Future investiga-
to conclude that, with respect to opera at least, good lieder per- tions should also include more excerpts for each performer so
formance will be sung with a more delayed vibrato onset, less that further comparisons can be made on a performer-by-
vibrato extent, and a weaker level of singers formant. Although performer level, rather than as a genre-specific analysis with
descriptions of what makes a good lied performance can encap- pooled degrees of freedom. However, even with these factors,
sulate the essence of these acoustic characteristics, until the pre- Studies 2 and 3 have cast light on quantitative differences in
sent research there was no evidence that could clarify and lied and opera singing styles and have demonstrated that
directly link such language to its acoustic counterparts. We although singing teachers have a good understanding of the
believe that bringing together the scientific and conventional required techniques that differentiate the styles, they may also
pedagogical approaches will help to provide a better under- benefit from examination and application of findings from
standing of the nature of singing and, here, has provided a better quantitative measurements in terms of SF and vibrato. What re-
understanding of the performance of lieder. mains to be seen is how the multitude of variables that consti-
The lack of consensus between the survey (study 1) and the tute human vocal production (we only examined a small
acoustic analyses (studies 2 and 3) in regards to vibrato onset selection, concerned with SF and vibrato) will provide a deeper
time may suggest that discrepancies are occurring from theoret- understanding of this fascinating and complex generator of
ical performance instruction to practical performance. There aesthetic pleasure. Additionally, the practical consideration of
are also variations among singers in the features they exploit accompaniment and environmental factors, such as piano
in the operatic context. For example, not all singers may want accompaniment in a small auditorium compared with a sym-
to or need to use a particularly strong SF, early vibrato onset, phony orchestra in a large auditorium, rather than genre alone,
and wide vibrato extent all of the time. Furthermore, although should receive further attention as the ontological basis of
some discrepancies were encountered, several descriptions of apparent differences between the two styles.
good lieder technique were complemented by the results of
the analyses. Among these, Elliott’s23 description of a ‘‘wide, REFERENCES
continuous vibrato’’ being undesirable in lieder is congruent 1. Sataloff RT. Professional Voice: The Science and Art of Clinical Care. New
with the findings of Studies 2 and 3. But care should be taken York: Raven Press; 1991.
not to assume that acoustic measurement provides a ground 2. Sundberg J. Level and center frequency of the singer’s formant. J Voice.
truth that informs pedagogical approaches. Methods are limited 2001;15:176–186.
3. Sundberg J, Romedahl C. Text intelligibility and the singer’s formant - a
in acoustic measurement contexts, too. For example, we had relationship? J Voice. 2009;23:539–545.
limitations on the kinds of notes we could analyze because of 4. Bretos J, Sundberg J. Measurements of vibrato parameters in long sustained
our decision to use commercial sound recordings and of those crescendo notes as sung by ten sopranos. J Voice. 2003;17:343–352.
to find notes that were not or weakly accompanied to allow ac- 5. Howes P, Callaghan J, Davis P, Kenny D, Thorpe W. The relationship be-
tween measured vibrato characteristics and perception in Western operatic
curate measurement. This approach could well have biased the
singing. J Voice. 2004;18:216–230.
sample of notes investigated, such as restricting the notes to 6. Sundberg J. My research on the singing voice from a rear-view perspective.
parts of an opera that were recitative or recitative-like and there- THM-QPRS. 2003;45:11–22.
fore perhaps more speech-like. We only examined the lower 7. Sundberg J, Gramming P, Lovetri J. Comparisons of pharynx, source,
voice types used in males. But confirmation of the effects of formant and pressure characteristics in operatic and musical theatre
vibrato is an area for further research for female vocalists. singing. J Voice. 1993;7:301–310.
8. Stone RE, Cleveland TF, Sundberg J, Prokop J. Aerodynamic and acous-
Furthermore, we identified a lack of consensus on how to tical measures of speech, operatic, and Broadway vocal styles in a profes-
measure the SF, having proposed three different methods in sional female singer. J Voice. 2003;17:283–297.
this research, which each produced slightly different results. 9. Bezerra A, Cukier-Blaj S, Duprat A, Camargo Z, Granato L. The character-
Of these, our analysis suggested that method C has good utility: isation of the vibrato in lyric and Sertanejo singing styles: acoustic and
It compares the level of the third formant against the level of the perceptual auditory aspects. J Voice. 2009;23:666–670.
10. Vurma A, Ross J. Where is a singer’s voice if it is placed ‘‘forward’’? J
first formant and although it produces some inaccuracies, it was Voice. 2002;16:383–391.
found to be overall conservative and relatively simple to calcu- 11. Kenny D, Mitchell H. Acoustic and perceptual appraisal of vocal gestures
late via visual inspection of a frequency spectrum. However, in the female classical voice. J Voice. 2006;20:55–70.
method A appeared to show good face validity because of the 12. Kenny D, and Mitchell H, ’Vocal quality in female classical singers: the role
of acoustics, perception and pedagogy’. In Proceedings of the 3rd confer-
phonetically plausible locations for selecting the two key read-
ence on Interdisciplinary Musicology, Tallinn, Estonia; 2007.
ings on the frequency spectrum for calculating LSF. 13. Ekholm E, Papagiannis G, Chagnin F. Relating objective measurements to
Even with such complications, we still found consistencies expert evaluation of voice quality in Western classical singing. J Voice.
across the qualitative and quantitative studies. Our point is 1998;12:182–196.
18 Journal of Voice, Vol. -, No. -, 2015

14. Bloothooft G, Plomp R. Spectral analysis of sung vowels: III characteristics 27. Wapnick J, Ekholm E. Expert consensus in solo voice performance evalu-
of singers and modes of singing. J Acoust Soc Am. 1986;79:852–864. ation. J Voice. 1997;11:429–436.
15. Mitchell H, Kenny D. The impact of ’open throat’ technique on vibrato rate, 28. Joliveau E, Smith J, Wolfe J. Vocal tract resonances in singing: the soprano
extent and onset in classical singing. Logoped Phoniatr Vocol. 2004;29: voice. J Acoust Soc Am. 2004;116:2343–2439.
171–182. 29. Audacity 1.3.3 Beta, multi-track audio editor and recorder, http://audacity.
16. Mitchell H, Kenny D. Open throat: acoustic and perceptual support for ped- sourceforge.net. GNU General Public License software.
agogic practice. J Singing. 2008;64:429–441. 30. Rossing TD, Sundberg J, Ternstr€om S. Acoustic comparison of voice use in
17. Diaz J, Rothman H. Acoustical comparison between samples of good and solo and choir singing. J Acoust Soc Am. 1986;79:1975–1981.
poor vibrato in singers. J Voice. 2003;17:179–184. 31. Omori K, Kacker A, Carroll L, Riley W, Blaugrund S. Singing power ratio:
18. Hakes J, Shipp T, Doherty E. Acoustic properties of straight tone, vibrato, quantitative evaluation of singing voice quality. J Voice. 1996;10:228–235.
trill and trillo. J Voice. 1987;1:148–156. 32. Lartillot O, Toiviainen P, Eerola T. ‘‘A matlab toolbox for music informa-
19. Sundberg J. Acoustic and psychoacoustic aspects of vocal vibrato. STL- tion retrieval,’’ in Data analysis, machine learning and applications. In:
QPSR. 1994;35:45–67. Preisach C, Burkhardt H, Schmidt-Thieme L, Decker R, eds. Berlin:
20. Weiss R, Brown W, Morris J. Singer’s formant in sopranos: fact or fiction. J Springer; 2008, pp.261–268.
Voice. 2001;15:456–468. 33. SPSS IBM, ’SPSS Statistics Package Version 22’. New York, U.S.A.: 2013.
21. Prame E. Measurements of the vibrato rate of ten singers. J Acoust Soc Am. 34. Sundberg J. The Science of the Singing Voice. Dekalb, Illinois: Northern Il-
1994;96:1979–1984. linois University Press; 1987.
22. Prame E. Vibrato extent and intonation in professional Western lyric 35. Schutte HK, Miller R. Intraindividual parameters of the singer’s formant.
singing. J Acoust Soc Am. 1997;102:616–621. Folia Phoniatr(Basel). 1985;37:31–35.
23. Elliot M. Singing in Style: A Guide to Vocal Performance Practices. Yale 36. Barnes JJ, Davis P. The relationship between professional operatic so-
University Press; 2006. prano voice and high range spectral energy. J Acoust Soc Am. 2004;116:
24. Spillman R. Performing lieder: the mysterious mix. In: Hallmark R, ed. 530–538.
German lieder in the nineteenth century. New York: Schirmer Books; 1996. 37. Sundberg J. The acoustics of the singing voice. Sci Am. 1977;236:82–91.
25. Miller R. Singing Schumann: An Interpretive Guide for Performers. New 38. Fletcher NH. ‘‘Vibrato in Music’’. Acoustics Australia. 2001;29:91–102.
York: Oxford University Press; 1999. 39. Bregman AS. Auditory Scene Analysis: The Perceptual Organization of
26. Finley R, and Finley C, ’SurveyMonkey’. California, U.S.A.; 2009. Sound. Massachusetts: MIT Press; 1990.