Predicting Mutational Change in the Speaking

Voice of Boys

*Michael Fuchs, †Matthias Fröehlich, ‡Bettina Hentschel, ||Ingo W. Stuermer,

|| Eberhard Kruse, and *Daniel Knauft

*‡Leipzig, †Erlangen, and kGoettingen, Germany

Summary: The authors investigated whether acoustic speaking voice analy-

ses can be used to predict the beginning of mutation in 21 male members of
a professional boys’ choir. Over a period of 3 years before mutation, children
were examined every 3 months by ear, nose, and throat (ENT) and phoniatric
specialists. At the same time, the voice was evaluated acoustically using anal-
ysis features of the Goettingen Hoarseness Diagram (GHD). Irregularity com-
ponent and noise component, jitter, shimmer, mean waveform correlation
coefficient, and fundamental frequency were determined from recordings
of the speaking voice. Significant changes of acoustic features appeared 7
and 5 months before mutation onset, which indicates that vocal function
is already restricted 6 months before mutation onset. This acoustic voice
analysis is therefore suitable to support the care of the professional singing
voice.
Key Words: Voice mutation—Acoustic analysis—Prediction—Goettingen
Hoarseness—Diagram—Adolescence.

INTRODUCTION vocal instability as, for instance, during mutation.1

Throughout Europe, numerous children and ado- Mutation is accompanied by a decreased vocal effi-
lescents sing in choirs. Some perform in profession- ciency caused by rapid growth of the larynx and the
al choirs with daily rehearsals and weekly concerts. vocal folds due to endocrine control of testosterone
Our clinical experience shows that young singers and growth hormones. In girls, the vocal folds be-
who are subject to high demands on their singing come 3–4 mm longer and the voicing fundamental
voice need a constant surveillance by singing teach- frequency (f0) drops by a third of an octave. In con-
ers and phoniatricians, especially during phases of trast, vocal folds in boys lengthen up to 10 mm

Accepted for publication October 25, 2005. Developmental and Voice Research Group, Goettingen,
Presented in part at the 18th Annual Meeting of the German Germany.
Society of Phoniatrics and Paedaudiology, 2001, Frankfurt/ Address correspondence and reprint requests to Dr. Michael
Main, Germany. Fuchs, University of Leipzig, Department of Otorhinolaryngol-
From the *University of Leipzig, Department of Otorhino- ogy, Liebigstrasse 18A, 04103 Leipzig, Germany. E-mail:
laryngology, Leipzig, Germany; the †Siemens Audiologische michael.fuchs@medizin.uni-leipzig.de
Technik GmbH Erlangen, Erlangen, Germany; the ‡University Journal of Voice, Vol. 21, No. 2, pp. 169–178
of Leipzig, Institute for Medical Informatics, Statistics and 0892-1997/$32.00
Epidemiology, Leipzig, Germany; and the ||University of Goet- Ó 2007 The Voice Foundation
tingen, Department of Phoniatrics and Pedaudiology, Sensory doi:10.1016/j.jvoice.2005.10.008

169
170 MICHAEL FUCHS ET AL

longer, whereas f0 drops by approximately 1 octave, mutation onset has to be predicted phoniatrically
which results in a higher sensitivity of the vocal before its symptoms are manifested.
apparatus during mutation in boys. The methods employed to diagnose mutation in
Whereas children with a normal level of vocal our clinic include video-stroboscopy and well-
strain usually pass through this period without se- proven analyses of voice efficiency and quality (eg,
vere problems, it is known that frequent singing voice range profile) that are performed repeatedly
during mutation can affect the physiological devel- during voice development. In previous studies, Hab-
opment of the voice.2 The changes in growth of the ermann8 and Bonet and Casan9 stressed the difficul-
vocal cords in the entire larynx described substan- ties entailed in distinguishing physiological from
tially increase vulnerability. The conditions during pathological dysphonia during mutation with partic-
mutation are often described with restrictions to ular reference to similarities in the manifestation of
the voice because of acute laryngitis.3 Moreover, mutation and mild laryngitis. Blood level testoster-
Seidner and Wendler4 point out that the neuromus- one, growth velocity, and genital status routinely
cular control during the growth phase is markedly were used to predict the remaining time up to muta-
restricted, which applies in particular to the singing tion or to establish whether it has occurred.1,10 Baken
voice. Young singers are endangered additionally and Orlikoff11 have already pointed out the need to
by reduced auditory self-control and a competitive identify morphological changes of the vocal appara-
attitude during choir rehearsals. Therefore, singing tus in its entirety with a view to evaluating mutational
teachers and phoniatricians prefer to exclude male effects on the voice by acoustic analyses.
singers from professional children’s choirs at the Pubertal voice change can be divided into three
onset of mutation. time periods: the premutation, the mutation as the
specific period of vocal instability, and the postmu-
Voice protection tation. Diagnosis and prediction of mutation are fo-
Voice protection presupposes prediction of the cused not on interindividual differences, but on
onset of voice mutation. Difficulties originate intraindividual changes. To our knowledge, so far
from the great individual variability and the gradual no one has investigated systematically whether
process of mutational voice changes, which make it computerized analyses of voice signals can provide
hard to define a clear cutoff point. Furthermore, on- the information required.
set of puberty and mutation accelerated over the Therefore, an acoustic analysis based on features
past century, although this phenomenon seems to taken from the Goettingen hoarseness diagram
have slowed down in recent decades.3,5 If studies (GHD) was applied to a longitudinal dataset of con-
on parameters of voice performance in children tinuing speaking voice signals from members of
and adolescents from the beginning of 1900, the a professional boys’ choir. The GHD was designed
1960s,3 the 1970s,6 and the 1980s7 are considered to distinguish between physiological and patholog-
and compared with today’s result, first of all, a con- ical voices and to describe the extent of deviation
tinuous advancement of physical development to an from a normal voice.12,13 The underlying acoustic
ever younger age is found. Since the 1970s, this features that have been investigated in this study
trend has continued, but very much more slowly were defined in such a way as to attain a stringent
than in the comparison period 1900–1970.5 Early description of voices over the complete range
mutation results in the attempt to retain boys’ so- from normal voice to complete aphonia.
prano/alto voices until just before mutation onset Computerized acoustic analyses can be applied
with the concomitant risk of voice dysfunction. to children and adolescents. Increasing efforts
Phoniatric research must establish first the cru- have been made to create normative databases of
cial time when vocal changes occur. This should young voices,14 and acoustic analyses were used
be defined as exactly as possible, and symptoms to describe voice production under pathological
of mutation must be distinguished from other phe- conditions, eg, functional voice disorders,15 vocal
nomena, eg, inflammation of the vocal apparatus fold nodules,16 in hearing-impaired children,17
or hyperactivity. Second, the time remaining until and postcochlear implantation.18

Journal of Voice, Vol. 21, No. 2, 2007

 PREDICTING MUTATIONAL CHANGE 171

The GHD enables an analysis method and graphic The GHD is highly sensitive to changes in pho-
representation of two of the most important inde- nation conditions along with a good reliability, as
pendent voice properties. The irregularity of the statistical analyses of data from different recording
acoustic signal is assessed by three acoustic param- sessions had shown.13 The GHD is an established
eters: jitter, shimmer, and mean waveform matching standard diagnostic tool in many European phoniat-
the correlation coefficient (MWC); all contribute ric departments and does not require expensive cus-
equally to the irregularity component (IC). Jitter tom hardware.
and shimmer are common features in the description The authors postulate that this acoustic analysis
of voice irregularities. Nevertheless, the terms do is suitable to provide evidence enabling differential
not always refer to the same physical properties. diagnosis of mutation and to predict the beginning
Usually, jitter is taken to refer to a variability in of the mutation.
fundamental frequency, whereas shimmer is associ-
ated with the variability in amplitude or energy on
a period-by-period basis. In the design of the METHODS
GHD, the highly variable algorithms were defined The authors examined 21 boys in a longitudinal
in such a way that their combination was optimal study between May 1992 and November 1995 until
in terms of several optimisation criteria (eg, stable the mutation had started for all subjects. Average
information content; see Michaelis et al12). In the age at the first examination was 11.2 years (9.3–
following, the terms ‘‘jitter’’ and ‘‘shimmer’’ always 12.5); average age at the beginning of mutation
comply with the particular definitions applied in the was 14.2 years (13.0–15.9). The respective voice
GHD. development stage (premutation, mutation) was es-
Besides irregularities, the additive noise content tablished on the basis of the subject’s history and
is one of the most important features. It is assessed the reports of the phoniatrician as well as the clin-
by the noise component (NC) that is based on the ical investigation including video-stroboscopy and
glottal-to- noise-excitation-ratio (GNE). A detailed the voice status. Apart from the symptoms de-
description is given by Michaelis et al.12,19 The scribed in video-stroboscopy, typical signs of muta-
GHD supplies an accurate description of different tion are a restriction of the pitch range and a change
voice generation mechanisms or laryngeal phona- in the sound of the voice with increase of the pro-
tion conditions.13 If the general laryngeal condition portions of roughness. The individual phases of
with regard to phonation is known, the GHD is suit- mutation can thus be appraised on the basis of the
able to define intraindividual voice changes as well clinical experience with several consecutive investi-
as interindividual group differences. It allows the gations at intervals of about 3 months.
comparison of female and male voices, because The last examination was carried out during the
none of its acoustic parameters depends on the de- mutation period. Each examination includes video-
termination of the fundamental frequency. stroboscopy and standardized voice analysis, eg,
In the GHD, the periodic parameters jitter, shim- voice range profile, and the recording of voice
mer, and MWC are calculated on the basis of the samples on digital audio tape (Denon DAT-
waveform matching algorithm.20,21 This algorithm recorder DRT-80, Denon, Pine Brook, NJ; Sennhe-
defines periods by cross-correlation of two adjacent iser stereo cond. microphone MKE-66, Sennheiser,
signal segments, and thus it works on such small Wedemark, Germany). The children were asked to
time scales that the word ‘‘frequency’’ cannot be read out the first verse of the hymn ‘‘Jesu meine
used directly to refer to a repetition of periods Freude.’’
over a certain time. However, the periodicity of Generally, automatic voice analysis systems rely
the waveform matching algorithm can be averaged on recordings of isolated vowels. Speech fluidity
over a longer stretch of time (‘‘frame’’) as a param- was used in the current study because isolated vow-
eter of the fundamental frequency. In this way, the els differ from the ‘‘normal’’ vocal use, and because
fundamental frequency was calculated for the voice subtle phonation changes before the beginning of
samples in this study. the mutation had to be detected and quantified.

Journal of Voice, Vol. 21, No. 2, 2007

172 MICHAEL FUCHS ET AL

FIGURE 1. Shimmer. Scatter plot of all data courses (n 5 21) from the first
examination to the beginning of the mutation (t 5 0), with a significant change
at 6 months before the beginning of the mutation (arrow).

A preprocessing segmentation algorithm is applied 19 input cells (one for each Bark channel), 6 hidden
that automatically selects the signal parts that cor- cells, and 1 output cell with a sigmoid characteristic
respond to voiced phonemes. These were later ana- was applied to classify the current frame into
lyzed acoustically (for details, see Lessing et al a voiced or an unvoiced phoneme.
22,23
). The spectral envelope was calculated by lin- During the calculation of the GHD, the acoustic
ear prediction, transformed to the Bark scale,24 and parameters jitter, shimmer, MWC, and GNE were
normalized for each time frame that did not belong calculated for each 40-ms signal frame that had been
to a speech pause. A neural network consisting of classified as voiced in the preprocessing segmentation

TABLE 1. Slope of Regression Lines and Their 95% Confidence Range for Each Parameter Derived From the
Examinations of 21 Boys Either Investigated #6 or O6 Months Before the Beginning of the Mutation
Slope (95% CI)

#6 Months O6 Months
Before Mutation Onset Before Mutation Onset

Irregularity component [skt/month] 0.0111 (0.0052; 0.0169) 0.0360 (0.0209; 0.0511)

Jitter [%/month] 0.0090 (20.0001; 0.0181) 0.1000 (0.0621; 0.1380)
Shimmer [%/month] 0.0641 (0.0294; 0.0989) 0.2772 (0.1689; 0.3855)
Mean waveform correlation coefficient [1/month] 20.0008 (20.0012; 20.0003) 20.0040 (20.0055; 20.0026)
Noise component [skt/month] 20.0006 (20.0067; 0.0055) 20.0330 (20.0506; 0.0154)

Journal of Voice, Vol. 21, No. 2, 2007

 PREDICTING MUTATIONAL CHANGE 173

FIGURE 2. Noise component. Scatter plot of all data courses (n 5 21) from the
first examination to the beginning of the mutation (t 5 0), with a significant
change at 6 months before the beginning of the mutation (arrow).

(provided that the complete continuous voiced seg- performed using SPSS version 9.0 (SPSS Inc.,
ment lasted at least 15 s). Based on these values, Chicago, IL) and S-Plus (Insightful, Durham, NC).
the IC and NC values were calculated. The data distri-
bution that resulted from all analyzed signal segments
of one recording session was characterized by the RESULTS
mean and the standard deviation for each coordinate Voice change occurred in all subjects before
and for each individual acoustic measure. In the November 1995. On average, 15 examinations
GHD, the means and standard deviations are used to were carried out in each boy (range 5–18, total
parameterize the data distribution.13,22,23 341). The standard GHD illustration in ellip-
All parameters were represented in scatter plots. tical form did not show any obvious correlation
Together with the singing teacher, the authors de- with the voice development status. Jitter, shimmer,
fined the onset of the mutation (t 5 0) under clini- MWC, and fundamental frequency were therefore
cal and heuristic aspects as the cutoff point for analyzed separately. The figures show the course
obvious changes of voice quality and efficiency of the individual parameters over time before muta-
over time, independent of the boy’s age. The au- tion and in relation to the beginning of mutation. Ir-
thors used curve fitting to identify significant respective of the age of the individual test subjects,
changes in the curves of each parameter retrospec- the time ‘‘0’’ of the x-axis was defined as the begin-
tively. Linear regression models were applied to re- ning of mutation. In this way, the curve charac-
peated measurements for the time before and after teristics in the period before mutation could be
the cut-point. All statistical analyses were observed. As a result of the curve fitting, each

Journal of Voice, Vol. 21, No. 2, 2007

174 MICHAEL FUCHS ET AL

FIGURE 3. Individual curve of jitter of test persons #19 and #25.

parameter showed a conspicuous change at a cutoff of 2.3 Skt (95% CI: 1.4; 3.2), and a mean waveform
point 6 months before the beginning of the muta- correlation of 0.9 (95% CI: 0.8; 1.0) are to be ex-
tion (t 5 0) (for shimmer and noise component, pected. After this cutoff point, irregularity compo-
see Figures 1 and 2). As the curves before and after nent, jitter, and shimmer show a sharp rise,
this cutoff point were characterized by approximate whereas noise component, mean waveform correla-
linear courses, the authors performed regression tion coefficient, and fundamental frequency
analyses for the respective datasets from the time decrease.
before and the time after this cutoff point. The Whereas the curves in Figures 1 and 2 show av-
slopes of the regression lines with 95% confidence erage values, individual analyses bear out the sig-
ranges are presented in Table 1. The confidence nificant changes at 6 months before the beginning
ranges of the two time intervals did not overlap. of mutation (eg, individual curves of two test per-
The authors can therefore conclude that the slopes sons; Figures 3 and 4). Thus, an individual predic-
differ significantly. In other words, each parameter tion of the time remaining until mutation is possible
changes its course significantly 6 months before it by the identification of the curve ‘‘jump.’’ The con-
is possible to diagnose the mutation by an auditive fidence range is exceeded 6 months before mutation
assessment or a clinical examination. (mentioned above), which facilitates interpretation
For the individual prediction of the time remain- of the curves.
ing until the onset of mutation, a fundamental fre-
quency of 226.2 Hz (95% CI: 187.8; 264.4), an
irregularity component of 6.3 Skt (95% CI: 5.5; CONCLUSIONS
7.1), a jitter of 1.6 % (95% CI: 0.3; 2.8), a shimmer The results confirm our hypothesis that acoustic
of 11.7 % (95% CI: 6.9; 16.5), a noise component analyses of the speaking voice are suitable to

Journal of Voice, Vol. 21, No. 2, 2007

 PREDICTING MUTATIONAL CHANGE 175

FIGURE 4. Individual curve of shimmer of test persons #19 and #25.

support the differential diagnosis of mutation and to Members of children’s and youth choirs usually
predict its onset. The 3-month schedule applied in change from soprano to alto during this phase at
our study allows changes of the voice quality dur- the latest.
ing voice development to be detected before they On the assumption that voice quality deteriorates
are perceptible on the basis of auditive assessment in the period after mutation, the irregularity compo-
or clinical methods. These analyses can support es- nent, noise component, jitter, and shimmer must be
tablished methods to predict the remaining time expected to increase, whereas the mean waveform
that a boy will be able to sing in soprano/alto, correlation coefficient decreases at the beginning
and to spare the singer’s daily vocal strain before of this period. All parameters showed a significant
onset of mutation. In this way, the method helps change in their courses as early as 6 months before
the conductor, the singing teacher, and the phonia- the mutation. This means all parameters change at
trician to provide make plans for every individual the same time before a loss of the voice quality is
boy’s voice in a choir.10 perceptible. All GHD-based parameters are inde-
The mean fundamental frequency decreases al- pendent of the fundamental frequency,12 so that
ready 6 months before the mutation during the pe- these observations cannot be attributed to the de-
riod of premutation. This correlates with crease in fundamental frequency that occurs simul-
physiological development, laryngeal growth, and taneously. For a reliable prediction of the time
a subsequent decrease of mean fundamental fre- remaining until mutation commences, several ex-
quency. The voice range changes periodically with- aminations are necessary during the premutation
out obvious effect on voice quality and efficiency.1 period and even the time before because the change

Journal of Voice, Vol. 21, No. 2, 2007

176 MICHAEL FUCHS ET AL

FIGURE 5. Scatter plot of all data courses (n 5 21) from the first examination
to the beginning of the mutation (t 5 0): lower limit of voice range.

of the course is discernible only by a documentation because the vocally trained child aims unconsciously
and comparison of at least three separate for a sound that is free of breathy and throaty
examinations. parts. It can be assumed that a gradual deterioration
Although this article was focused on speaking of the voice quality can thus be compensated to
voice, the authors also found noticeable changes some extent to conceal the approaching mutation.1
of singing voice, eg, a reduction of the upper and On the other hand, one cannot rule out an effect of
lower limit of voice range within the voice range the algorithm that calculates the glottal-to-noise ex-
profile in corresponding figures (Figure 5). On the citation ratio due to the small basis of the current
other hand, there was no significant change in the data. Even though the glottal-to-noise is in principle
voice range. independent of the fundamental frequency,
At the moment, a predictive statement and a dif- a systematic change caused by a decreasing funda-
ferential diagnosis are only possible retrospectively. mental frequency may lead to a slight systematic
In this study, the data pool was too small to create trend.
normative data for boys. The authors suggest that Irrespective of the cause of the decrease in the
acoustic analyses should be included in regular ex- noise component, the data show that all measures
aminations of boys’ voices. may be used to assess initial morphological
The noise component shows a significant de- changes accurately, as well as the resulting limita-
crease starting 6 months before the onset of muta- tion of vocal efficiency and, possibly, vocal
tion instead of an expected increase that may be compensation.
explained as follows: On the one hand, it seems The results confirm our hypothesis that acoustic
plausible that compensation mechanisms may analyses using the GHD are suitable to provide ev-
occur with regard to the changes in vocalization idence for differential diagnosis of the mutation

Journal of Voice, Vol. 21, No. 2, 2007

 PREDICTING MUTATIONAL CHANGE 177

and for prediction of its beginning. The informa- Further studies are required to establish whether
tion that nonaudible changes of voice occur 6 other changes of the singing voice can be verified
months before the beginning of mutation in the with the GHD. Reference values for GHD in child-
GHD permits reliable evaluation of the time re- hood and adolescence are required.
maining until mutation as well as a differential di-
agnostic statement on the presence of the mutation.
This supplements and enhances the phoniatric ex- REFERENCES
amination and other established methods that pre- 1. Fuchs M, Behrendt W, Keller E, Kratzsch J. Methods for
dict the time of the mutation. This method can be prediction of the beginning of mutation in boys voices:
combined, eg, with determination of the level of investigations in singers of the Thomaner Choir Leipzig.
testosterone in the blood that enables relatively ex- Folia Phoniatr Logop. 1999;51:261–271.
2. Biesalski P, Frank F. Phoniatrie–Pädaudiologie. Vol. 1.
act prediction of the time remaining up to mutation
Stuttgart, Germany: Thieme; 1994:297–298.
onset.1 The determination of the growth rate is also 3. Frank F. Mutation in consideration of acceleration. Heilpä-
appropriate to appraise the voice development be- dagogik. 1969;12:18–20.
fore beginning of mutation when the lowest point 4. Seidner W, Wendler J. The Singer’s Voice. Berlin: Hent-
of the curve of the growth rate is determined at schelverlag Kunst und Gesellschaft; 1982:175–177.
the beginning of puberty.10 Simultaneous determi- 5. Schilling A, Karthaus K. Acceleration and mutation. HNO.
1961;9:271–275.
nation of these parameters and the interpretation 6. Frank F, Sparber M. New evidence of voice ranges in chil-
of the results together with the acoustic analyses de- dren. Folia Phoniatr Logop. 1970;22:397–402.
scribed in this article are of great assistance in the 7. Doll J, Frank F, Sparber M. New evidence of vocal devel-
appraisal of voice development by the phoniatrician opment in 12- to 16-years old boys. Sprache Stimme Ge-
and the voice teacher.25,26 Further studies are re- hör. 1986;10:143–146.
8. Habermann G. Voice and Speech–An Introduction in its
quired to show whether the statistical reliability
Physiology and Hygienics. Stuttgart, Germany: Thieme;
can be enhanced by combining the individual 1986:146–147,247,256.
techniques. 9. Bonet M, Casan P. Evaluation of dysphonia in a children’s
The main advantage of the new method con- choir. Folia Phoniatr Logop. 1994;46:28.
sists in its noninvasive character. Furthermore, 10. Fuchs M, Behrendt W, Kratzsch J, Keller E. Forecast of
it does not impose any strain on the children. voice mutation at singers of professional boy’s choirs
with parameters of growth and puberty, insulin-like growth
GHD does not require any expensive custom hard- factor I and testosterone—investigation on singers of the
ware and can be used in daily clinical practice. Thomanerchor Leipzig. Horm Res. 1997;48:133.
However, a prerequisite for a proper interpreta- 11. Baken RJ, Orlikoff RF. Voice measurement: is more
tion is that the acoustic analyses are applied to better? Log Phon Vocol. 1997;22:147–151.
a longitudinal dataset.1 The concept of an interdis- 12. Michaelis D, Fröhlich M, Strube HW. Selection and com-
bination of acoustic features for the description of patho-
ciplinary management to safeguard professional
logic voices. J Acoust Soc Am. 1998;103:1628–1639.
voices in childhood and adolescence includes a pe- 13. Fröhlich M, Michaelis D, Strube HW, Kruse E. Acoustic
riodic phoniatric examination in periods of vocal voice analysis by means of the hoarseness diagram.
instability. The new method can be integrated J Speech Lang Hear Res. 2000;43:706–720.
into these investigations. The authors recommend 14. Campisi P, Tewfik TL, Manoukian JJ, Schloss MD, Pel-
examinations at intervals of about 3 months, which land-Blais E, Sadeghi N. Computer-assisted voice analysis
establishing a pediatric database. Arch Otolaryngol Head
should include phoniatric examination includ- Neck Surg. 2002;128:156–160.
ing video-stroboscopy, voice status including voice 15. Niedzielska G. Acoustic analysis in the diagnostic of voice
range profile, physical examination including gen- disorders in children. Int J Ped Otorhinolaryngol. 2001;57:
ital status and growth rate, measurement of the 189–193.
capillary blood level of testosterone, and acoustic 16. Niedzielska G, Glijer E, Niedzielski A. Acoustic analysis
of voice in children with noduli vocales. Int J Ped Otorhi-
analyses using GHD. All methods can be com-
nolaryngol. 2001;60:119–122.
bined depending on the particular question under 17. Monsen RB. Acoustic qualities of phonation in young
examination and the results of the preceding hearing-impaired children. J Speech Hear Res. 1979;22:
investigations. 270–288.

Journal of Voice, Vol. 21, No. 2, 2007

178 MICHAEL FUCHS ET AL

18. Szkiekowska A, Maniecka-Aleksandrowicz B, Dolecki J. Witte H, eds. Proceedings of the 4th International Work-
Voice rehabilitation in children post cochlear implantation. shop. University of Jena; 2000:124–131.
Central East Eur J Otorhinolaryngol Head Neck Surg. 23. Lessing J, Fröhlich M, Michaelis D, Strube HW, Kruse E.
2000;5:23–27. Use of neuronal networks for assessment of voice quality in
19. Michaelis D, Gramss T, Strube HW. Glottal-to-noise exci- pathological voices. In: Gross M, ed, Aktuelle Phoniatrisch-
tation ratio—a new measure for describing pathological Pädaudiologische Aspekte 1998/99. Vol. 6. Heidelberg,
voices. Acta Acustica. 1997;83:700–706. Germany: Median; 1999:39–43.
20. Milenkovic P. Least mean square measures of voice pertur- 24. Zwicker E, Fastl H, Frater H. Psychoacoustics: Facts and
bation. J Speech Hear Res. 1987;30:529–538. Models. Vol. 22. Berlin: Springer; 2000.
21. Titze IR, Liang H. Comparison of F0 extraction methods for 25. Behrendt W, Fuchs M. Mutation of singing voice in adoles-
high-precision voice perturbation measurements. J Speech cence. In: Gundermann H, ed. Die Ausdruckswelt der Stimme.
Hear Res. 1993;36:1120–1133. 1st ed. Heidelberg, Germany: Hüther; 1998:122–127.
22. Lessing J, Fröhlich M, Michaelis D, Strube HW, Kruse E. 26. Fuchs M, Behrendt W. Diagnostical value of voice range
A neural network based method to assess voice quality profile for phoniatrical care of professional singing voice
from continuous speech for different voice disorders. Ad- during mutation. In: Gross M, ed. Aktuelle Phoniatrisch-
vances in quantitative laryngosocopy, voice and speech re- Pädaudiologische Aspekte1999/2000. Vol. 7. 1st ed. Hei-
search. In: Braunschweig T, Hanson J, Schellhorn-Neise P, delberg, Germany: Median; 2000:133–138.

Journal of Voice, Vol. 21, No. 2, 2007