Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=ilog20

Download by: [78.13.230.161] Date: 18 May 2016, At: 05:41

Logopedics Phoniatrics Vocology

ISSN: 1401-5439 (Print) 1651-2022 (Online) Journal homepage: http://www.tandfonline.com/loi/ilog20

Freddie Mercury—acoustic analysis ofspeaking fundamental frequency, vibrato, andsubharmonics

Christian T. Herbst, Stellan Hertegard, Daniel Zangger-Borch & Per-ÅkeLindestad

To cite this article: Christian T. Herbst, Stellan Hertegard, Daniel Zangger-Borch &Per-Åke Lindestad (2016): Freddie Mercury—acoustic analysis of speaking fundamentalfrequency, vibrato, and subharmonics, Logopedics Phoniatrics Vocology, DOI:10.3109/14015439.2016.1156737

To link to this article: http://dx.doi.org/10.3109/14015439.2016.1156737

View supplementary material

Published online: 15 Apr 2016.

Submit your article to this journal

Article views: 107081

View related articles

View Crossmark data

ORIGINAL ARTICLE

Freddie Mercury—acoustic analysis of speaking fundamental frequency, vibrato,and subharmonics

Christian T. Herbsta, Stellan Hertegardb, Daniel Zangger-Borchc and Per-Åke Lindestadb

aVoice Research Lab, Department of Biophysics, Faculty of Science, Palack�y University Olomouc, 17. listopadu 12, 771 46 Olomouc, CzechRepublic; bDepartment of Otolaryngology, Head and Neck Surgery B53, Clinical Sciences and Intervention, Karolinska Institutet, KarolinskaUniversity Hospital, 141 86 Stockholm, Sweden; cVoice Centre Stockholm AB, Zangger Vocal Art, Box 4342, 102 67 Stockholm, Sweden

ABSTRACTFreddie Mercury was one of the twentieth century’s best-known singers of commercial contemporarymusic. This study presents an acoustical analysis of his voice production and singing style, based onperceptual and quantitative analysis of publicly available sound recordings. Analysis of six interviewsrevealed a median speaking fundamental frequency of 117.3 Hz, which is typically found for a baritonevoice. Analysis of voice tracks isolated from full band recordings suggested that the singing voice rangewas 37 semitones within the pitch range of F#2 (about 92.2 Hz) to G5 (about 784 Hz). Evidence forhigher phonations up to a fundamental frequency of 1,347 Hz was not deemed reliable. Analysis of 240sustained notes from 21 a-cappella recordings revealed a surprisingly high mean fundamental fre-quency modulation rate (vibrato) of 7.0 Hz, reaching the range of vocal tremor. Quantitative analysis uti-lizing a newly introduced parameter to assess the regularity of vocal vibrato corroborated itsperceptually irregular nature, suggesting that vibrato (ir)regularity is a distinctive feature of the singingvoice. Imitation of subharmonic phonation samples by a professional rock singer, documented by endo-scopic high-speed video at 4,132 frames per second, revealed a 3:1 frequency locked vibratory patternof vocal folds and ventricular folds.

ARTICLE HISTORYReceived 4 September 2015Revised 18 November 2015Accepted 12 February 2016Published online 14 April2016

KEY WORDSFreddie Mercury; Queen;singing voice range;speaking fundamentalfrequency; subharmonics;ventricular folds; vibrato;voice quality

Introduction

Freddie Mercury, born 5 September 1946 as Farrokh Bulsara,and died 24 November 1991, was one of the most influentialsingers of his time. The Rolling Stone magazine listed himeighteenth within the 100 best (commercial contemporary)singers of all time (1). Having been the lead singer of therock band Queen, he profoundly influenced this group’smusical style over more than two decades.

Freddie Mercury’s voice has been described as ‘a force ofnature with the velocity of a hurricane’ (2), which was ‘esca-lating within a few bars from a deep, throaty rock-growl totender, vibrant tenor, then on to a high-pitched, perfect col-oratura, pure and crystalline in the upper reaches’ (3, p. 2).Such descriptions, while presumably adequate for a biog-raphy or a newspaper article, do not satisfy deeper scholarlyinterest into the singer’s voice characteristics. The purpose ofthis study was therefore to conduct a viable analysis of pub-licly available data material, in order to arrive at moreempirically based insights into Freddie Mercury’s voice pro-duction and singing style.

Materials and methods

The task of objectively describing the voice of a singer whois not available for acquiring recordings under controlled

acoustical and physiological conditions is ambitious. Specialcare needs to be taken in selecting proper data material(ideally a cappella, no background sounds) and suitable ana-lysis methods. In this study, a series of exemplary interpreta-tions of voice quality were conducted, and, where deemedappropriate, quantitative estimation of voice features likespeaking fundamental frequency or vibrato properties wasperformed. These analyses, in part inspired by a previouspublication by the first author (4), are conducted here withimproved methodology and larger data sets. Finally, endo-scopic high-speed video examination of vocal fold vibrationof a contemporary rock singer mimicking typical sounds pro-duced by Freddie Mercury was performed, in an attempt tounderstand the voice production mechanism of subharmonicsounds. No formant analysis was undertaken, however, sinceit cannot be taken for granted that the spectral contents ofcommercially available recordings have not been considerablychanged by the editing sound engineers.

Speaking fundamental frequency (SFF) and singing voicerange

In order to assess Freddie Mercury’s speaking fundamentalfrequency (SFF), a total of six interviews, conducted byAndrew Wigg (1984, 1985, 1986, 1987) and Rudi Dolezal

CONTACT Christian T. Herbst herbst@ccrma.stanford.edu Voice Research Lab, Department of Biophysics, Faculty of Science, Palack�y University Olomouc, 17.listopadu 12, 771 46 Olomouc, Czech Republic

Supplemental data for this article can be accessed here

� 2016 Informa UK Limited, trading as Taylor & Francis Group

LOGOPEDICS PHONIATRICS VOCOLOGY, 2016http://dx.doi.org/10.3109/14015439.2016.1156737

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

(1984 and 1987) were analyzed. The interviews by AndresWigg were taken from the commercial compilation FreddieMercury: The Solo Collection (2000, EMI/Parlophone). Theinterview conducted by Andrew Wigg in 1979 was excludedfrom the analysis due to excessive background noise. Theinterviews by Rudi Dolezal (‘A musical prostitute’, 1984, and‘The last interview’, 1987) were downloaded from the inter-net (https://www.youtube.com/watch?v¼8wk9hPubD1Q andhttps://www.youtube.com/watch?v¼PnsiSjsSNYQ).

The acoustic tracks of all interviews were normalized inamplitude and then annotated and analyzed with the soft-ware Praat (5). Only those segments were considered wherewithout doubt Freddie Mercury’s voice was exclusively aud-ible. The fundamental frequency of all extracted sequenceswas estimated using Praat’s autocorrelation algorithm (6),using a rather conservative parameter setting in order toavoid analysis artifacts stemming from background noise orunvoiced consonants: fundamental frequency range:70–400 Hz; silence threshold: 0.2; voicing threshold: 0.6; oct-ave cost: 0.01; octave jump cost: 0.8; voiced/unvoiced cost:0.14; read progress: 10 ms. The analysis resulted in a total of134,418 data points.

For the perceptual assessment of the singing voice range,Freddie Mercury: The Solo Collection as well as 23 commer-cially available recordings of Queen were considered. Theassessment was conducted by author C.T.H. by matching theperceived pitch of candidate sound samples, sung atextremely high and low frequencies, to the perceived musicalpitch of a piano tuned to A4¼ 440 Hz (equal temperament).Only those sound samples were considered where FreddieMercury could be clearly identified as the exclusively audiblesinger, and no background vocals were included in theanalysis.

Descriptive analysis of functional singing voiceparameters

In speech and singing, the spectral composition of the emit-ted sound is causally determined by the configuration of thesound production organ, constituted by the respiratory sys-tem, the larynx, and the vocal tract. At the laryngeal level,two main determinants are constituted by the vocal registerand the degree of posterior vocal fold adduction, controlledalong the dimension of ‘breathy’ versus ‘pressed’ (7,8).

Without any physiological or biomechanical data of vocalfold vibration the assessment of the laryngeal configurationduring singing is difficult, particularly when based on com-mercial recordings whose spectral composition could havebeen changed by sound engineers and by influences of therecording equipment (microphone and amplifier transferfunctions, data conversion, recording device), room acoustics(reverberation), and the mouth-to-microphone distance(high-frequency selective directionality) (9). Nevertheless, asmall number of exemplary sound samples are discussedhere, based on perceptual and (partly) spectral analysis, andhypothetical physiological interpretations as regards vocalregister and posterior vocal fold adduction are suggested.The perceptual assessments were conducted by the four

authors of this manuscript (two graduated voice pedagoguesand two phoniatricians), and all assessments are based onunanimous consensus of the four authors.

Analysis of vocal vibrato

Vibrato is constituted by a (predominantly regular) modula-tion of the singing fundamental frequency, typically in therange of 4–7 Hz (10). For the assessment of FreddieMercury’s vibrato characteristics, the compilation TheAcapella Collection (http://queenspain.blogspot.co.at/2013/06/freddie-mercury-acapella-collection.html), compiled frommultichannel recordings by the Queen Poland fan site(https://queenpoland.wordpress.com/), was analyzed. Thisassortment of studio recordings contains the vocal tracks of28 songs by Queen and Freddie Mercury. Sustained notesfrom these recordings have been considered for analysiswhen the following conditions were met: 1) exactly one sing-ing voice was audible, and this voice could be clearly identi-fied as Freddie Mercury’s; 2) no background noise,background vocals, or orchestration was audible; 3) no note-worthy delay or reverberation effects have been applied; and4) the sustained note had a duration of more than 500 msand contained audible cyclic frequency modulations.

Data annotation and f0 extraction was conducted with thesoftware Praat. For f0 extraction, Praat’s autocorrelationalgorithm was used with these parameters: timeStep:10 ms; frequency range 70–600 Hz; voicingThreshold: 0.3;octaveJumpCost: 0.35; silenceThreshold: 0.03; octaveCost:0.07; voicedUnvoicedCost: 0.14. After the automated analysis,all extracted fundamental frequency contours were superim-posed upon the respective spectrograms and then visuallyassessed for correct f0 extraction. Nine segments wereexcluded from the analysis since they contained obviouserrors (mostly octave jump errors during the occurrence ofsubharmonics).

A total of 128 segments of sustained notes (from 15 of the28 songs in The Acappella Collection) with vibrato (11,475data points total, 114.75 seconds total duration), having reli-ably estimated f0 contours of over 500 ms duration each, wereconsidered for further analysis. For each segment, three quan-tities were computed: 1) the average fundamental frequencyf 0 ; 2) the average absolute deviation from the mean musicalpitch Dc as an estimator of vibrato extent; and 3) the domin-ant modulation frequency fmod (see further below).

Vibrato extent (i.e. the modulation amplitude) is typicallyexpressed in semitones or cents, where one octave equals 12semitones or 1,200 cents (11). For the purpose of this ana-lysis, the frequency values of the fundamental frequency con-tours were converted to cents relative to middle C as

c½i� ¼ 1200logð f0½i�

f0 C4Þ

logð2Þ (1)

where f0[i] is the fundamental frequency estimate at timeindex i, and f0_C4 is the fundamental frequency of middle C(C4), calculated as 440� 2̂(–9/12) � 261.63 Hz. The averagemusical pitch c of the f0 contour, expressed in cents from

2 C. T. HERBST ET AL.

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

middle C, was then computed as

c ¼ 1n

Xn�1

i¼0

c½i� (2)

Finally, the average absolute deviation from the meanmusical pitch Dc, representing vibrato extent, was computedas

Dc ¼ 1n

Xn�1

i¼0

c � c½i�j j (3)

As a reference, for completely regular vibrato with per-fectly sinusoidal frequency modulation, the vibrato extentwould equal twice the (maximum) modulation amplitudedivided by p, i.e. the amplitude divided by about 1.57.Consequently, a perfectly sinusoidal vibrato with a modula-tion amplitude of 61 semitone would correspond to aboutDc � 63.7 cents. In order to rule out analysis artifacts, foursegments with Dc < 127.3 (representing a perfectly sinus-oidal vibrato with an extent of more than two semitones)were excluded from the analyzed data set.

A completely regular vibrato naturally contains only onemodulation frequency (see e.g. Figure 1A for an example ofLuciano Pavarotti’s vocal vibrato with an almost sinusoidalmodulation frequency of about 5.7 Hz—example 1.1 fromMiller (12)). In contrast, preliminary inspection of FreddieMercury’s vocal vibrato suggested more irregular frequencymodulation patterns, caused by the superposition of morethan one modulation frequency component (see Figure 1Bfor an example, taken from ‘Bohemian Rhapsody’, t ¼ 71 sin The Acapella Collection release (t ¼ 132 s in the A Nightat the Opera album), ‘go’ within the phrase ‘Good-bye every-body, I’ve got to go’).

When considering the f0 contour in Figure 1B it becameapparent that time domain methods (like peak-picking) fordetermining the number of vibrato cycles per second areproblematic: it is unclear whether the small ripples around t¼ 1 should be counted as individual cycles or not. Therefore,in order to obtain a robust numerical parameter for assessing

Freddie Mercury’s vibrato, the dominant modulation fre-quency (fmod) was computed with Fourier analysis. For eachsegment, the DC offset (i.e. the mean value) of each f0 con-tour was removed. The normalized f0 contour was scaled bya Hann window, and then zero-padded to eight times its ori-ginal length. The resulting time series vectors were subjectedto a forward FFT, and the amplitude spectrum was calcu-lated. The frequency of the spectral maximum in the ampli-tude spectrum was considered as the dominant modulationfrequency, fmod. In order to rule out artifacts, four segmentswith fmod < 3 Hz were excluded from the analyzed data set.

As a by-product of the FFT analysis of the f0 contours,the amplitude difference of the two strongest frequency com-ponents of the f0 modulation contour was calculated asDAFmod ¼ (1 – A2)/A1, where A1 and A2 are the two stron-gest components found in the normalized amplitude spec-trum of the f0 contour (dark triangles and light diamonds,respectively, in Figure 1A and B). The DAFmod parameter isan estimator of the vibrato regularity. It can theoreticallyassume values in the range of 0 to 1, where 0 would repre-sent f0 modulation with the two strongest frequency compo-nents in the f0 contour being equally strong, and 1 wouldrepresent purely sinusoidal frequency modulation with onesingle modulation frequency component. When consideringthe preliminary analysis done in Figure 1, the respectiveDAFmod values found in Luciano Pavarotti’s and FreddieMercury’s vibrato are about 0.89 and 0.11, suggesting a quiteregular vibrato for Luciano Pavarotti and an irregular vibratowith two almost equally strong modulation frequency com-ponents for Freddie Mercury.

Imitation and analysis of rough voice production

Preliminary perceptual analysis suggested that a portion ofthe available data material was sung with a pronounceddegree of roughness. The phenomenon was most clearly aud-ible in sustained notes found at t ¼ 31 s, 39 s, 120 s (‘on’within the phrase ‘turn it on’) in the a-cappella track ‘Let’s

(A)

(B)

Figure 1. Illustration of vibrato characteristics analysis. Preliminary data for two sustained notes sung by Luciano Pavarotti (A) and Freddie Mercury (B), see text. Leftpanels: fundamental frequency contour; right panels: normalized amplitude spectrum of fundamental frequency contour computed by Fourier analysis (see Methods).The two strongest spectral components within the range of 2–13 Hz (light gray vertical lines) are marked by a dark triangle and a light diamond, respectively, andtheir relative amplitude difference DAFmod is indicated by the dashed line.

LOGOPEDICS PHONIATRICS VOCOLOGY 3

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

Turn It On’ from Freddie Mercury: The Solo Collection(2000, EMI/Parlophone). The respective sound exampleswere extracted and played back to a professional rock singer(co-author D.Z.-B.) who served as the subject in previousstudies (13,14). The singer was asked to imitate these sounds,and phonation was recorded with a 70� rigid endoscope(Karl Storz, Tuttlingen, Germany) attached to a Hispec 1b/w high-speed camera (Fastec Imaging, San Diego, CA,USA) which was operated at 4,132 frames per second, with a300 W xenon light source (Richard Wolf GMBH,Knittlingen, Germany). The video recordings were stored ona computer as uncompressed AVI files. Digital kymograms(DKGs) (15,16) were extracted with a Jython plugin (http://homepage.univie.ac.at/christian.herbst/index.php?page¼fiji)written by author C.T.H. for the Fiji image analysis frame-work (17).

The acoustic signal was recorded with a Sennheiser MKE-2 microphone (Sennheiser, Wennebostel, Germany) mountedon a headset at 15 cm distance from the singers’ mouth,using a custom-built microphone amplifier. The sound wasdigitized with a LSI LSI PC/C32 sound card (LoughboroughSound Images plc, Loughborough, Leicestershire, England) at

a sampling frequency of 16,000 Hz. The digitized sound wasstored in the smp file format using the software Soundswelland Phog (Electronix Hitech, T€aby, Sweden), and was laterconverted to and stored as uncompressed wav.

Results

Speaking fundamental frequency

Fundamental frequency data extracted from the analyzedinterviews was not normally distributed. Overall, an averagespeaking fundamental frequency of 130.1 Hz was computed,with a median at 117.3 Hz and the 5th and 95th percentilesat 89.1 Hz and 231.4 Hz, respectively. Boxplots of the funda-mental frequency extracted from the six interviews are shownin Figure 2. The mean and median values are provided inTable 1. In order to test whether the interviewer had aneffect on the speaking fundamental frequency, all data pointswere pooled by interviewer. Interviews conducted by RudiDolezal had a mean speaking fundamental frequency of119.3 Hz, with the median at 111.2 Hz and the 5th and 95thpercentiles at 87.3 and 174.5 Hz, respectively. The interviews

(A)

(B)

Figure 2. A: Interquartile range distributions of speaking fundamental frequency estimates for the six analyzed interviews. The whiskers indicate the 5th and the 95thpercentile, respectively, and the superimposed stars indicate the group means. B: Histogram of all speaking fundamental frequency data, grouped by interviewer(Dolezal: dark gray; Wigg: light gray; Cumulative: white).

Table 1. Average and median speaking fundamental frequency values (including 5th and 95th percentile) for the six analyzed interviews.

Year Interviewer Number of data points Average f0 (Hz) Median f0 (Hz) 5th percentile f0 (Hz) 95th percentile f0 (Hz)

1984 Dolezal 26,124 119.2 111.1 87.1 175.91984 Wigg 39,487 142.2 125.6 96.5 253.41985 Wigg 11,795 130.4 121.5 93.7 187.31986 Wigg 22,772 120.9 107.2 86.2 218.71987 Dolezal 13,816 119.6 111.4 87.6 173.21987 Wigg 20,424 137.7 124.8 88.3 240.5

4 C. T. HERBST ET AL.

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

conducted by Andrew Wigg had generally higher speakingfundamental frequency values, with an average of 134.6 Hz,the median being at 120.2 Hz and the 5th and 95th percen-tiles at 90.0 Hz and 245.7 Hz, respectively. A Kruskal–Wallisrank sum test performed on the non-normally distributeddata with the software R (18) revealed a highly significantdifference between the two groups (Dolezal versus Wigg,P< 0.001).

Singing voice range

The lowest note found in the analyzed data corpus was ashort F#2 (about 92.2 Hz) in ‘Don’t Try Suicide’ (The Game),found at t ¼ 52.5 s (‘so’ within the phrase ‘So you think it’sthe easy way out?’). A series of sustained notes at musicalpitch G2 (about 98 Hz) were found in ‘Get Down MakeLove’ at t ¼ 58.7 s, 61.8 s, 116.7 s, 120 s, 146.7 s, and 149.9 sin the version from The Acapella Collection (‘down’ withinthe phrase ‘Ev’ry time I get hot you wanna cool down’)—note that in relation to the version on the News of the Worldalbum, these time offsets are shifted by about 12.5 seconds.For the highest notes a couple of candidates were found: 1) asustained E5 (about 659.3 Hz) in ‘Hang on in There’ (TheMiracle) at t ¼ 134.5 s (‘wait’ within the phrase ‘wait for thatmoment’); 2) a sustained F5 (about 698.5 Hz) in ‘All God’sPeople’ (Innuendo) at t ¼ 42 s (background voice ‘yeah’,right after ‘majesty around the world’), which was masked byother sung sounds in the recording; 3) a very short butclearly perceivable G5 (about 784 Hz) in ‘Let’s Turn It On’(Freddie Mercury: The Solo Collection) at t ¼ 25 s (‘on’

within the phrase ‘let’s turn it on’), which was most probablyproduced in falsetto register.

Descriptive analysis of functional singing voiceparameters

Overall, based on perceptual assessment, Freddie Mercuryseemed to have ample control over vocal registration and‘blending the registers’, i.e. mixing the chest and falsettoregisters). More often than not, both these registers tendedto have comparable timbral characteristics. Sounds producedin the falsetto register, where unambiguously identifiable (e.g.‘Let’s Turn It On’ at t ¼ 23.2–25.2 s), could contain har-monic content up to 10 kHz and sometimes even beyond.(Note, though, that any existing harmonic content could havebeen amplified by the sound engineer during the post-pro-duction stage of the recording.) A few examples where thechoice of vocal registration was more evident due to abrupttransitions between the registers include: 1) ‘Nevermore’(Queen II), t ¼ 15–28 s (‘Can’t you see? Listen to the breeze,whisper to me please. Don’t send me to the path of never-more’); 2) ‘Seaside Rendezvous’ (A Night at the Opera), t ¼34–45.2 s (‘Can we do it again? Can we do it again some-time? (ooh I like that) Fantastic, c’est la vie Madames etMonsieurs’); 3) ‘Love of My Life’ (A Night at the Opera), t ¼131–134.6 s (‘How I still love you’), where there was a per-ceivable and abrupt timbral shift between ‘love’ (sung in chestregister) and ‘you’ (falsetto register); or 4) ‘Keep Passing theOpen Windows’ (The Works) at t ¼ 18–23 s, where ‘need’ ofthe phase ‘love is all you need’ again bore perceptual clues offalsetto voice production.

Figure 3. Examples of ‘breathy’ and ‘pressed’ sound production. A and B: narrow-band spectrograms (FFT bin width 10.77 Hz, dynamic range 55 dB); C and D: corre-sponding amplitude spectra, extracted at t¼ 20.5 s and t¼ 2 s, respectively. The light gray (orange color online) line indicates the noise floor at about –40 dB and–55 dB, respectively.

LOGOPEDICS PHONIATRICS VOCOLOGY 5

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

Freddie Mercury appeared to have had good control overthe phonation type along the dimension ‘breathy’ to‘pressed’. Two such examples illustrating the range of timbralpossibilities are illustrated in Figure 3. The example forbreathy phonation (Figure 3A and C), taken from ‘TeoToriatte’ (A Day at the Races) contained clearly audiblebroadband noise above 1.5 kHz, which could hypotheticallyhave been created by turbulent noise (19), presumably causedby incomplete vocal fold adduction. This broadband noisewas strong enough clearly to highlight several vocal tract res-onances, identified as the second, third, and fourth formantin the spectrogram (see arrow annotation in Figure 3Aand C). An example for the ‘pressed’ phonation type is illus-trated in Figure 3B and D. Apart from the perceptuallystrained quality of the sound and the apparent lack of vocalvibrato, the example was characterized by a clearly audiblerelease of subglottal pressure at the end of the phrase (t ¼5.2 s). When the normalized spectra of the two antagonisticexamples are compared, clear differences of spectral compos-ition become apparent. The ‘breathy’ example, although pol-luted by keyboard accompaniment, contained a far smallernumber of noteworthy harmonics, and the normalized spec-trum had a higher noise floor (about –40 dB overall) as com-pared to the ‘pressed’ example (about –55 dB)—see light gray(orange color online) lines in Figure 3C and D.

Vibrato

Boxplots of the calculated vibrato parameters are shown inFigure 4. The average fundamental frequency of the analyzedvibrato notes was 350.2 Hz, with a median of 347.0 Hz and5th and 95th percentiles at 199.6 Hz and 494.9 Hz, respect-ively. The mean dominant modulation frequency fmod was7.0 Hz (median 7.2 Hz, 5th and 95th percentiles at 5.2 Hzand 8.2 Hz). The average vibrato extent Dc was calculated tobe 35.0 cents (corresponding to a perfectly sinusoidal vibratowith an amplitude of about 55 cents or 0.55 semitones—seeMethods), with a median of 32.1 cents and the 5th and 95thpercentiles at 19.6 cents and 64.4 cents, respectively. Thevibrato regularity estimator DAFmod had a mean value of0.57, with a median at 0.60 and 5th and 95th percentiles at0.09 and 0.97, respectively. The variation of dominant modu-lation frequency fmod and modulation extent rC as a function

of fundamental frequency were analyzed with linear regres-sion analysis. While there was only a slight tendency fordecreased dominant modulation frequency as f0 increases(fmod ¼ –0.0033 f0 þ 8.19, R2 ¼ 0.09), there existed a slightlystronger (but still weak) inverse correlation between the f0and modulation extent (Dc ¼ –0.0674 f0 þ 58.98, R2 ¼0.21). In other words, the vibrato extent tended to be smallerat higher pitches.

Subharmonic voice production

An example of Freddie Mercury’s phonation with a perceiveddegree of roughness is illustrated in Figure 5A. In the spec-trogram, the harmonics appear as integer multiples of thefundamental frequency, and so-called subharmonics emergeat integer multiples of a third of the fundamental frequency,constituting a period-3 subharmonic pattern (period tri-pling). The same phenomenon is seen in the vocal imitation

(A) (B) (C) (D)

Figure 4. Interquartile range distributions of computed vibrato parameters. The whiskers indicate the 5th and the 95th percentile, respectively. A: Fundamental fre-quency of analyzed notes sung with vibrato; B: dominant modulation frequency fmod; C: vibrato extent; D: vibrato regularity measure DAFmod.

Figure 5. Narrow-band spectrograms of subharmonic voice production. A:Freddie Mercury: ‘Let’s Turn It On’ (The Acapella Collection), t¼ 67 s; B: Imitatedsound sung by professional rock singer D.Z.-B.

6 C. T. HERBST ET AL.

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

produced by professional rock singer D.Z.-B. (Figure 5B).Analysis of the laryngeal endoscopic HSV data from D.Z.-B.’s phonation revealed frequency-locked vibration of theventricular folds at a third of the frequency of vibration ofthe vocal folds (see Supplementary Movie M1, availableonline). This was also seen in the corresponding DKGs(Figure 6), extracted at two positions along the anterior–pos-terior vocal fold dimension. Anteriorly, the ventricular foldvibration exhibited clear lateral surface waves (similar tomucosal waves within the true vocal folds), possibly inducedby high levels of glottal air flow and subglottal pressure.Overall, the epilaryngeal tube was constricted, thus limitingthe visibility of the true vocal folds. The arytenoid cartilages(and thus also the vocal processes) were firmly adducted.

Discussion

The main difficulty of this retrospective study was posed bythe fact that no controlled data acquisition was possible. Theresults of the acoustic and perceptual analyses conducted hereneed thus be treated with care: not all sound samples were a-cappella recordings, variable recording tape wheel speedscould have slightly changed the measured fundamental fre-quencies, and any spectral component could have beenchanged by the intervention of sound engineers at the post-production stage (therefore no formant data were analyzed).Except for the vocal imitation by rock singer D.Z.-B. nophysiological data were available to back up the physiology-related conclusions drawn from the analyzed acoustic data. Inspite of these complications, this study was successful in deliv-ering a few insights into Freddie Mercury’s voice productionand singing style. These are summarized and discussed below.

Speaking voice fundamental frequency

Analysis of interviews revealed a mean SFF of 130.1 Hz(median at 117.3 Hz). Morris et al. found an average SFF of127.4 Hz (610.1 Hz) for trained middle-aged baritones andbasses (20); Brown et al. documented an average SFF of119 Hz (no standard deviation given) for the same group

(21); and Chernobel’skii reported a SFF of 123.2 Hz(610.2 Hz) for baritones (22). Based on these data it can behypothesized that Freddie Mercury had a baritone’s speakingvoice, since the respective mean SFF values for trained tenorswere about 20 Hz higher (20–22). This assumption is furthermotivated by the notion that the SFF is located at a roughlypredictable offset within a speaker’s total f0 range (23).Normative values given for that offset in various studies arein the range of 3–8 semitones above the lowest achievablefundamental frequency (20,24,25), and this variability mightbe caused by effects of vocal training (23). Anecdotally, thesuggestion that Freddie Mercury had a baritone’s voice rangeis corroborated by a quote from his duet partner, classicalsoprano Montserrat Caball�e, who maintained that ‘He had abaritone voice. I told him one day, ‘‘Let’s do a small duet ofbaritone and soprano’’, and he said, ‘‘No, no, my fans onlyknow me as a rock singer and they will not recognise myvoice if I sing in baritone’’’ (2).

There was a noteworthy and statistically significant differ-ence between SFF data from interviews by Andrew Wigg ver-sus Rudi Dolezal. Interviews conducted by Andrew Wiggcontained several spoken phrases where the SFF exceeded200 Hz, with maximum values up to 320 Hz (which occurredduring connected speech as a means of putting emphasis oncertain words) as indicated by manual inspection of the rawdata material. Other confounding factors notwithstanding,the SFF difference might be attributed to Andrew Wiggbeing a native speaker of British English (as was FreddieMercury), thus creating a more familiar and personal atmos-phere during the interviews that allowed for a somewhathigher SFF. In contrast, the Austrian Rudi Dolezal was anative speaker of German.

Singing voice range

Measurement of the singing voice range constituted the mostchallenging task in this study, since it was solely based onperceptual assessment of commercially available recordings.The determined extreme pitches of F#2 (about 92.5 Hz) andG5 (about 784 Hz) constitute a singing voice range of 37

(B)

(A) (C) (D)

Time (82 ms total)

right

left

right

left

vocal foldvibration

vibration of rightventricular fold

ventricular foldlateral surface waves

Figure 6. Endoscopic evidence of subharmonic vocal fold vibration. A and B: digital kymograms (DKGs); C and D: depiction of virtual kymographic scan lines for theDKGs shown in A and B, respectively.

LOGOPEDICS PHONIATRICS VOCOLOGY 7

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

semitones. Such values are well in line with observations byMorris et al., who report an average voice range of 38.7semitones for middle-aged baritones (20).

Online sources like Wikipedia (26) report singing voiceranges of four octaves from F2 (about 87.3 Hz) to F6 (about1,396.9 Hz) for Freddie Mercury. However, in the exampleslisted by these sources the vocalist is not clearly identifiable(mostly because the putative examples for extreme notesstem from background vocals, which could have been sungby other members of Queen), or the respective data sup-posedly came from vocal improvisations that were not avail-able as data material in this study. A sustained note with amaximum fundamental frequency of 1,347 Hz (slightly abovemusical pitch E6) was found in ‘It’s Late’ (News of theWorld) at t ¼ 370–372 s. On perceptual grounds it is unclearwhether this note was sung by Freddie Mercury or any othermember of Queen, or whether it was played by a guitar.While such high fundamental frequencies of notes werereported to be possible for a ‘male soprano’ (27) and formales using the ‘whistle register’ (28), it can, given the ana-lyzed data material, not be reliably concluded (nor can it beruled out) that such a phenomenon existed in FreddieMercury’s voice.

Descriptive analysis of functional singing voiceparameters

Evidence for describing the functional aspects of FreddieMercury’s singing habits was derived from perceptual assess-ment and spectral analysis of selected sound samples. Thepresented analysis (see Results and Figure 3) suggested theuse of different vocal registers (termed ‘chest’ and ‘falsetto’ inthis manuscript), as well as different phonation types alongthe dimension of ‘breathy’ versus ‘pressed’. However, as nophysiological data (i.e. endoscopic laryngeal recordings, etc.)were available, no rigorous proof can be given. The assump-tions made in this study vis-�a-vis laryngeal configurationsand physiological quantities, such as subglottal pressure (seeFigure 3 and corresponding text), remain therefore specula-tive. This weakness is mainly constituted by the most obvi-ous fact that this work is a post hoc analysis of a singer whowould not be available for further data acquisition.

Nonetheless, based on the perceptual impression of theassessed recordings it is not too far-fetched to conjecture thatFreddie Mercury was rather skillful in adapting his laryngealconfiguration to musical needs, thus exhibiting a great vari-ability of sound timbres for enhanced musical expression.

Vibrato

For Western classical singing, vibrato typically has a rate ofabout 5.4 Hz to 6.9 Hz (29–31). This rate may depend on ageand emotional involvement (29), and on esthetical preferen-ces which may have changed over the last 100 years (32).For non-classical singing, values of 5.45 Hz (33), 6.09 Hz(34), and 5.04 Hz (35) were reported. In contrast, the averagedominant frequency of Freddie Mercury’s fundamental fre-quency modulation was found to be 7.04 Hz, which is about

1 to 2 Hz higher than average vibrato rates reported for non-classical singers, thus being more in line with modulationrates reported for vocal tremor (36,37). As such, the ten-dency for higher fundamental frequency modulation rates inFreddie Mercury’s voice may well contribute to his uniquevocal style.

The average vibrato extent Dc of 35.38 cents (correspond-ing to a perfectly sinusoidal vibrato with an amplitude ofabout 55 cents or about 0.55 semitones) is by and large com-parable to the average values previously reported for non-clas-sical singing, which range from 30.9 cents (reported in percentand converted to cents by C.T.H.) in singers of ‘contemporarymusic, pop, gospel and jazz’ (33) to 66.78 cents for male jazzsingers (34). These values are generally lower than the vibratoextent reported for classical singing, ranging from 71 cents(30) to 112 in ‘Lied’ and 138 cents in operatic singing (31).

In a perfectly sinusoidal vibrato, the DAFmod parameter, asdefined in this study, assumes a value of 1. A value very closeto 1 has been found for operatic tenor Luciano Pavarotti inthe preliminary analysis (see Methods and Figure 1). In con-trast, the analysis of Freddie Mercury’s notes sung withvibrato resulted in an average DAFmod value of 0.57, and halfthe analyzed samples had values of 0.6 or smaller, indicatingthat largely more than one modulation frequency componentwas found in the spectral analysis of the f0 contours.

The principal vibrato agent is assumed to be a muscle ora group of muscles that regulates fundamental frequency.Typically, the cricothyroid is hypothesized to form anagonist-antagonist pair with the thyroarytenoid and/or thelateral cricothyroid muscles (10,38). In the ‘standard’ case ofregular vibrato (see e.g. the Pavarotti example in Figure 1),such an agonist–antagonist system would exhibit simplesinusoidal oscillation. Such a simple model could, however,not explain the complex f0 modulation seen in some ofFreddie Mercury’s sustained notes. It is thus hypothesizedthat either a single agonist–antagonist pair exhibited complexirregular oscillation, or more than one agonist–antagonistpairs for fundamental frequency modulation were present.Naturally, as no physiological data were available, neitherposition can be maintained. The proposed notion might,however, be relevant for further physiological investigationinvolving contemporary singers.

Perceptually, Freddie Mercury’s irregular (and typicallyfaster) vibrato is clearly audible in the sustained notes offamous songs such as ‘Bohemian Rhapsody’ (A Night at theOpera) or ‘We Are the Champions’ (News of the World), andit appears to be one of the hallmarks of his vocal style. On amore general level, these findings suggest that the regularityof vibrato (indicated here by the DAFmod parameter) could bean important feature for any singing voice, which maydeserve to receive further scientific attention.

Subharmonic voice production

The analysis of rock singer D.Z.-B.’s imitation of FreddieMercury’s ‘rough’ singing style revealed subharmonic vibration(39,40) created by 3:1 frequency locking of the vocal folds andthe ventricular folds: during each vibration of the ventricularfolds the vocal folds completed three oscillations. Similar

8 C. T. HERBST ET AL.

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

vibratory patterns of the vocal and the ventricular folds assynchronized oscillators were observed in previous studiesinvolving the same (14) and other singers (41,42). Theobserved lateral surface waves along the ventricular folds arecomparable to mucosal waves in true vocal folds (43,44).

This type of voice production, involving vibratingventricular folds at various degrees, has been termed‘growl’ (45), ‘dist’ (14), ‘throat-singing’ (46), or ‘vocal-ven-tricular mode’ (VVM) (42). The umbrella term ‘distortion’(47) or ‘intentional distortion’ (ID) (48) has been suggestedfor singing voice effects where, in addition to the ventricularfolds, also other supraglottal structures are engaged invibration.

Spectrograms of sound production by both D.Z.-B. andFreddie Mercury show comparable subharmonic patterns(period tripling). It is therefore likely that frequency-lockedventricular fold vibration also occurred in the original sampleof Freddie Mercury’s ‘rough’ singing which was imitated in thisstudy. An alternative, but less likely explanation would suggestsubharmonic oscillation of the vocal folds alone (49), or a fre-quency-locked vibration of the vocal folds and the aryepiglotticfolds (45).

Voice production and stage persona

Subharmonics, as described in the previous paragraph, are apossible route of a system ‘on its way to chaos’ (50), withpotential communicative relevance (51). Their occurrenceaids in creating the impression of a sound production systemdriven to its limits, even while used with great finesse. Thesetraits, in combination with the fast and irregular vibrato,might have helped create Freddie Mercury’s eccentricand flamboyant stage persona. This stage persona and itsapparent constant concourse with Freddie Mercury’s moreprivate side is best characterized by some quotes from hisinterviews: ‘I’m just a musical prostitute’ (Dolezal, 1984); ‘Iam the nicest person you could meet, my dear. It’s justthat I am handicapped, and therefore I have to sort of virtu-ally fight my so-called stage persona. Most of the times itsort of works against me’ (Wigg, 1984); and, concerning theaudience’s expectations and losing control: ‘Because of mypersona on stage they think that I carry on this way off-stage. If I did, I would have been dead a long time ago’(Wigg, 1984).

Acknowledgements

This study is a work of ‘fan-science’. Author contributions: C.T.H. con-ceived the study, collected the data material, conducted the perceptualand quantitative analysis, and wrote the manuscript. S.H., D.Z.-B. andP.A.L. collected the vocal imitation data and contributed to the finalversion of the manuscript. We are very grateful to Dr Hans Larsson(Karolinska University Hospital) for his help in acquiring the respectiveendoscopic data. Parts of this manuscript were previously published in:Christian T. Herbst (2012). Freddie Mercury—Akustische Stimm-Analyse. L.O.G.O.S. Interdisziplin€ar, 20 (3), 174–83.

Declaration of interest

The authors report no conflicts of interest.

Funding information

This study was partially financed by the institutional fund of thePalack�y University Olomouc, Czech Republic (C.T.H.).

References

1. 100 Greatest Singers: Freddie Mercury. New York, NY: RollingStone; 2011 [cited 2011 Nov 14]; Available from: http://www. roll-ingstone.com/music/lists/100-greatest-singers-of-all-time-19691231/.

2. Sullivan C. Freddie Mercury: the great enigma. London, UK: TheGuardian; 2012. Available from: http://www.theguardian.com/music/2012/sep/27/freddie-mercury-great-enigma.

3. Bret D. Living on the edge: the Freddie Mercury Story. London,UK: Robson Books, 1996.

4. Herbst CT. Freddie Mercury—Akustische Stimmanalyse. LOGOSInterdisziplin€ar. 2012;20:174–83.

5. Boersma P, Weenink D. Praat: doing phonetics by computer.Amsterdam, The Netherlands: Institute of Phonetic Sciences,University of Amsterdam, 2014.

6. Boersma P, editor. Accurate short-term analysis of the fundamen-tal frequency and the harmonics-to-noise ratio of a sampledsound. Proceedings of the Institute of Phonetic Sciences, 1993.Amsterdam; 1993.

7. Herbst CT, Howard DM, Svec JG. The sound source in singing –basic principles and muscular adjustments for fine-tuning vocaltimbre. In: Welch G, Howard DM, Nix J, editors. The Oxfordhandbook of singing. Oxford, UK: Oxford University Press, inpress.

8. Herbst CT, Hess M, M€uller F, Svec JG, Sundberg J. Glottal adduc-tion and subglottal pressure in singing. J Voice 2015;29:391–402.

9. Svec JG, Granqvist S. Guidelines for selecting microphones forhuman voice production research. Am J Speech Lang Pathol2010;19:356–68.

10. Hirano M, Hibi S, Hagino S. Physiological aspects of vibrato. In:Dejonckere PH, Hirano M, Sundberg J, editors. Vibrato SanDiego: Singular Publishing Group, 1995. p. 9–33.

11. Young RW. Terminology for logarithmic frequency units.J Acoust Soc Am 1939;11:134–9.

12. Miller DG. Resonance in singing Princeton, NJ: Inside ViewPress, 2008.

13. Zangger Borch D, Sundberg J. Some phonatory and resonatorycharacteristics of the rock, pop, soul, and Swedish dance bandstyles of singing. J Voice 2010;25:532–7.

14. Zangger Borch D, Sundberg J, Lindestad PA, Thalen M. Vocalfold vibration and voice source aperiodicity in ‘dist’ tones: a studyof a timbral ornament in rock singing. Logoped Phoniatr Vocol2004;29:147–53.

15. Svec JG, Schutte HK, Sram F. Videokymography: high-speed linescanning of vocal fold vibration. In: McCafferty G, Coman W,Carroll R, editors. XVI World Congress of OtorhinolaryngologyHead and Neck Surgery, Sydney ‘97 Proceedings, vol 2. Bologna:Monduzzi Editore, 1997. p. 1685–8.

16. Wittenberg T, Tigges M, Mergell P, Eysholdt U. Functional imag-ing of vocal fold vibration: digital multislice high-speed kymogra-phy. J Voice 2000;14:422–42.

17. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M,Pietzsch T, et al. Fiji: an open-source platform for biological-image analysis. Nat Methods 2012;9:676–82.

18. R: A language and environment for statistical computing. In:Team RDC, editor. Vienna, Austria: R Foundation for StatisticalComputing, Vienna, Austria, 2011.

19. Childers D, Lee CK. Vocal quality factors: analysis, synthesis, andperception. J Acoust Soc Am 1991;90:2394–410.

20. Morris RJ, Brown WS Jr, Hicks DM, Howell E. Phonational pro-files of male trained singers and nonsingers. J Voice 1995;9:142–8.

21. Brown W, Morris RJ, Hollien H, Howell E. Speaking fundamentalfrequency characteristics as a function of age and professionalsinging. J Voice 1991;5:310–15.

LOGOPEDICS PHONIATRICS VOCOLOGY 9

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016

22. Chernobel’skii SI. [The speaking fundamental frequency and thesinging voice type]. Vestn Otorinolaringol 2010(6):36–7. [Russian]

23. Baken RJ, Orlikoff RF. Clinical measurement of speech and voice2nd ed. Singular Publishing, Thompson Learning; 2000.

24. Awan SN. Superimposition of speaking voice characteristics andphonetograms in untrained and trained vocal groups. J Voice1993;7:30–7.

25. Friedrich G, Dejonckere PH. Das Stimmdiagnostik�Protokollder European Laryngological Society (ELS)—erste Erfahrungen imRahmen einer Multizenterstudie. Laryngo-, Rhino-, Otologie2005;84:744–52.

26. Wikipedia. Freddie Mercury. Wikimedia Foundation; 2015 [cited1 June 2015]; Available from: http://en.wikipedia.org/wiki/Freddie_Mercury.

27. Castrato Directed by Francesca Kemp. London: BBC Four, 2006.28. Edgerton M. Pitch profile of the glottal whistle. Malaysian

Journal of Science 2013;32:78–85.29. Sundberg J. Acoustic and psychoacoustic aspects of vocal vibrato.

In: Dejonckere PH, Hirano M, Sundberg J, editors. Vibrato SanDiego: Singular Publishing Group, 1995. p. 35–62.

30. Prame E. Vibrato extent and intonation in professional Westernlyric singing. J Acoust Soc Am 1997;102:616–21.

31. Johnson-Read L, Chmiel A, Schubert E, Wolfe J. PerformingLieder: expert perspectives and comparison of vibrato and singer’sformant with opera singers. J Voice 2015;29:645.e15–32.

32. Titze IR. Principles of voice production. National Center forVoice and Speech, 2000.

33. Reddy AA, Subramanian U. Do rate and extent of vibrato varywith gender in untrained singers? International Journal ofInterdisciplinary Research and Innovations 2014;2:1–5.

34. Manfredi C, Barbagallo D, Baracca G, Orlandi S, Bandini A,Dejonckere PH. Automatic assessment of acoustic parameters ofthe singing voice: application to professional Western operaticand jazz singers. J Voice 2015;29:517.e1–9.

35. Pecoraro G, Curcio DF, Behlau M, editors. Vibrato rate variabilityin three professional singing styles: opera, rock and Braziliancountry. Proceedings of Meetings on Acoustics; 2013. Montreal,CA: Acoustical Society of America, 2013.

36. Ramig L, Shipp T. Comparative measures of vocal tremor andvocal vibrato. J Voice 1987;1:163–7.

37. Hertegard S, Granqvist S, Lindestad PA. Botulinum toxin injec-tions for essential voice tremor. Ann Otol Rhinol Laryngol2000;109:204–9.

38. Titze IR, Story B, Smith M, Long R. A reflex resonance model ofvocal vibrato. J Acoust Soc Am 2002;111(5 Pt 1):2272–82.

39. Titze IR, Baken RJ, Herzel H. Evidence of chaos in vocal foldvibration. In: Titze IR, editor. Vocal fold physiology: frontiers inbasic science San Diego, CA: Singular Publishing Group, 1993. p.143–88.

40. Herzel H, Berry D, Titze IR, Saleh M. Analysis of vocal disorderswith methods from nonlinear dynamics. J Speech Hear Res1994;37:1008–19.

41. Bailly L, Henrich N, Pelorson X. Vocal fold and ventricular foldvibration in period-doubling phonation: physiological descriptionand aerodynamic modeling. J Acoust Soc Am 2010;127:3212–22.

42. Fuks L, Hammarberg B, Sundberg J. A self-sustained vocal-ven-tricular phonation mode: acoustical, aerodynamic and glotto-graphic evidences. Royal Institute of Technology - Speech, Musicand Hearing Quarterly Progress and Status Report (Stockholm).1998;3/1998:49–59.

43. Hirano M. Clinical examination of voice New York: Springer-Verlag, 1981.

44. Berke G, Gerratt BR. Laryngeal biomechanics: an overview ofmucosal wave mechanics. J Voice 1993;7:123–8.

45. Sakakibara K-I, Fuks L, Imagawa H, Tayama N, editors. Growlvoice in ethic and pop styles. Proceedings of the InternationalSymposium on Musical Acoustics (ISMA2004); 2004. Nara,Japan, 2004.

46. Lindestad PA, Sodersten M, Merker B, Granqvist S. Voice sourcecharacteristics in Mongolian ‘‘throat singing’’ studied with high-speed imaging technique, acoustic spectra, and inverse filtering. JVoice 2001;15:78–85.

47. Titze I. Deliberate use of distortion in singing. In: Kuhl P, CrumL, editors. 16th International Congress on Acoustics and 135thMeeting Acoustical Society of America: Vol I The Sound of theFuture: A Global View of Acoustics in the 21st Century. Seattle:University of Washington, 1998. p. 435–6.

48. McGlashan J, Sayles M, Kjelin H, Sadolin C, editors. Vocal effectsin singing: a study of intentional distortion using laryngostrobo-scopy and electrolaryngography. AQL2013, 10th internationalconference ‘‘Advances in quantitative laryngology, voice andspeech research’’; 2013. Cincinnati, OH: AQL Press, 2013.

49. Herzel H, Berry D, Titze I, Steinecke I. Nonlinear dynamics ofthe voice: signal analysis and biomechanical modelling. Chaos1995;5:30–4.

50. Lauterborn W, Cramer E. Subharmonic routes to chaos observedin acoustics. PhysRevLett 1981;47:1445–8.

51. Fitch WT, Neubauer J, Herzel H. Calls out of chaos: the adaptivesignificance of nonlinear phenomena in mammalian vocal pro-duction. Animal Behaviour 2002;63:407–18.

10 C. T. HERBST ET AL.

Dow

nloa

ded

by [

78.1

3.23

0.16

1] a

t 05:

41 1

8 M

ay 2

016