Long-Term Average Spectrum in Screening of Voice

Quality in Speech: Untrained Male

University Students
Timo Leino

Summary: Voice quality has mainly been studied in trained speakers, singers, and dysphonic patients. Few studies
have concerned ordinary untrained university students voices. In light of earlier studies of professional voice users,
it was hypothesized that good, poor, and intermediate voices would be distinguishable on the basis of long-term average
spectrum characteristics. In the present study, voice quality of 50 Finnish vocally untrained male university students was
studied perceptually and using long-term average spectrum analysis of text reading samples of one minute duration.
Equivalent sound level (Leq) of text reading was also measured. According to the results, the good and ordinary voices
differed from the poor ones in their relatively higher sound level in the frequency range of 13 kHz and a prominent peak
at 34 kHz. Good voices, however, did not differ from the ordinary voices in terms of the characteristics of the long-term
average spectrum (LTAS). The strength of the peak at 34 kHz and the voice-quality scores correlated weakly but sig-
nificantly. Voice quality and alpha ratio (level difference above and below 1 kHz) correlated likewise. Leq was signif-
icantly higher in the students with good and ordinary voices than in those with poor voices. The connections between
Leq, voice quality, and the formation of the peak at 34 kHz warrant further studies.
Key Words: Voice qualityLTAS.

INTRODUCTION discriminate satisfactorily between various articulatory set-

Sound quality is defined as the attribute of sound that makes
two sounds of the same pitch and loudness dissimilar.1,2 By
definition, then, quality, loudness, and pitch are distinct percep-
tual attributes even though a change in one may affect another.
The physical correlate of sound quality is sound energy distribu-
tion along the frequency range. Long-term average spectrum
(LTAS) analysis provides a means of viewing the average fre-
quency distribution of the sound energy in a continuous speech
sample. The method yields information of the general voice
quality, if the duration of the sample is long enough to avoid
the effects of individual speech sounds. Li et al3 regard a sample
of 3040 seconds long enough to study voice quality, whereas
Majewski et al4 recommend a sample of at least 1 minute.
LTAS has been used to study differences between normal and
pathological voices58 and between various vocal patholo-
gies.913Among normal-voiced subjects LTAS has been used
to study individual and gender, age and language related differ-
ences.4,1420 Characteristics of vocal expression of emo-
tions21,22 and acoustic differences between specific voice
qualities2325 have also been studied with this method. More-
over, LTAS has been used to investigate singing voice, for exam-
ple, differences between voice categories and various singing
styles.26,27
The suitability of LTAS for vocal screening has also been
questioned. In the study by Kitzing,24 the differences between
the LTAS of perceptually very different voice qualities were in
the magnitude of only 3%6%. Nor did Nolan find LTAS to
Accepted for publication March 26, 2008.
From the Department of Speech Communication and Voice Research, University of
Tampere, Tampere, Finland.
Address correspondence and reprint requests to Timo Leino, Department of Speech
Communication and Voice Research, University of Tampere, Kalevantie 4, FIN-33014
Tampere, Finland. E-mail: timo.leino@uta.fi
Journal of Voice, Vol. 23, No. 6, pp. 671-676
0892-1997/$36.00
2009 The Voice Foundation
doi:10.1016/j.jvoice.2008.03.008
tings.23 He concluded that LTAS may show better glottal
than resonatory characteristics of voice quality. Lofqvist et
Mandersson28 found substantial differences in the LTAS spec-
tra recorded from the same subjects during the same day,
which they supposed to impair the usefulness of LTAS for
screening purposes. However, their recordings were made before
and after a working day requiring a lot of speaking. Substantial
differences in LTAS have been reported after vocal loading.2931
Encouraging results on the suitability of LTAS for differenti-
ation of various voice qualities have been reported for both
normal and disordered voices.11,13,25
To obtain some objectivity in setting goals for vocal training
and for evaluating its effects, Leino used LTAS to study the dif-
ferences between good and poor speaking voices of vocally
trained professional male actors.32,33According to the results
LTAS distinguished between good, fairly good, rather poor,
and poor voice qualities. Poor speaking voices had relatively
lower sound level in the frequency range of 13 kHz, and
good voice quality differed from fairly good and rather poor
voice qualities by having a more prominent peak between 3
and 4 kHz. This peak seemed to be a kind of an actors formant
(ref. singers formant34). Perceptual evaluation of voice quality
correlated with the level of the peak.
The present study investigates the voice quality of normal-
voiced, vocally untrained male university students.
MATERIALS AND METHODS
Subjects and recording
Text reading samples of 50 Finnish male normal-voiced vocally
untrained university students (2027 years, mean 22 years; no
acting students included) were recorded in the same sound-
treated studio using a digital recorder and a Bruel and Kjaer mi-
crophone (4165) at a distance of 40 cm from the subjects
mouth. The speakers were asked to read the same prose extract
in their habitual way, using natural loudness, that is, neither too
672
soft nor with extra effort, and with neutral expression, thus, no
artistic performances were allowed. No external monitoring
was provided to control for F0 or sound level, because this would
have impaired the naturalness of the speech. The recordings
were perceptually monitored by the researchers, and the subjects
were asked to repeat the task if some of the requirements re-
ported above were not met. The duration of each sample was
over one minute. The samples were calibrated for sound pres-
sure level measurements using a sine wave generator and a level
meter (Bruel and Kjaer Frequency Analyzer 2120).
Listening evaluation
The reading samples were evaluated by two experienced profes-
sional voice trainers and five students of vocology. The listeners
rated the voice quality using a scale 3, 2, 1, 0, +1, +2, +3
(3 very poor, +3 excellent). All subjects had training in
perceptual analysis of voices, which had been given in the
same department. According to the terminology used in the per-
ceptual training, voice quality is one of the three basic character-
istics of voice together with pitch and loudness. Listening tests
were carried out in a sound-treated studio. The samples were
played back with a high-quality loudspeaker (Genelec Biamp
1019A). The listeners participated in the test in two groups to al-
low each subject to be seated sufficiently close (at 2 m) to the
loudspeaker, directly in front of it. The whole 1-minute samples
were presented. Leq was set to be the same in each sample on the
listening tape. The Intelligent Speech Analyser (ISA) signal
analysis system, developed by Raimo Toivonen, M.Sc. Eng.,
was used in this standardization. First the true Leq of each sam-
ple was measured. Then one of the samples was chosen as the
reference and the Leq of all the other samples were set up to
be the same as that of the reference by increasing or decreasing
their level by the amount needed.
Spectrum analysis
LTAS was made with a real-time spectrum analyzer (Hewlett-
Packard signal analyzer 3561A). A 400-point linear narrow
band fast Fourier transform (FFT) analysis was used. Earlier
findings33 have shown that broadband analysis does not differ-
entiate sufficiently between voice qualities.
The frequency span was 10 kHz; fast display mode was
used since it provides with real-time registration up to
7.5 kHz. Voiceless segments were left out of the analysis
by using an analogous s-gate. Pauses were excluded by using
the amplitude trigger of the analyzer. The time record length
was 40 millisecond, and the overlap was 15%. The display
resolution was 25 Hz. The flat top window with the fre-
quency bandwidth of 98 Hz was used because it is practically
ripple free (<0.01 dB) in the passband, and thus gives the
best amplitude accuracy. This characteristic makes Flat Top
window suitable, for instance, for calibration purposes.
The LTAS were normalized for the strongest peak and com-
pared to each other according to the level differences between
frequency ranges. The fact that relative levels were compared
instead of absolute levels made the subjects comparable to
each other regardless of the absolute Leq values of their sam-
ples. In addition, the level difference between the first formant
Journal of Voice, Vol. 23, No. 6, 2009
region and the fundamental frequency range was measured by
the subtraction L1L0 (see Figure 1). This level difference
has been found to correlate negatively with the perception of
breathiness (hypofunctional voice production35). That level dif-
ference may also be affected by intensity and F0. In addition to
LTAS measurements, alpha ratio9 was also calculated with In-
telligent Speech Analyser by subtracting the Leq of the range of
50 Hz1 kHz from that of the range of 15 kHz.
The text reading samples were divided into different voice
quality classes according to the points given in the listening
evaluation (poor mean points 3 to 1.2, good: +0.8to
+2.6, intermediate: 0.8 to +0.4). The statistical significance
of the differences between the voice quality groups was studied
by carrying out Students unpaired t-tests and nonparametric
MannWhitney U tests. To illustrate the results, the LTAS of in-
dividual samples were averaged over voice quality classes with
a custom-made computer program (developed by Heikki Ala-
talo, DSP-Systems).
RESULTS
Test results
Interrater and intrarater reliability of the listeners evaluations
was satisfactory (Cronbachs alpha 0.89, Spearmans Corre-
lation r 0.75, P < 0.001, respectively). There was a positive
correlation between the points given for voice quality and the
average sound energy level in different frequency ranges
(Spearman correlation coefficient r 0.39, P 0.005 for 1
2 kHz, r 0.40, P 0.004 for 23 kHz, and r 0.33,
P 0.018 for 34 kHz). L1L0 difference did not correlate
significantly with the perceptual evaluation. Alpha ratio also
correlated with voice quality (r 0.31, P 0.032).
The average points given in the listening evaluation allowed
a grouping of the voice samples into 14 good voices, 14 poor
voices, and 22 intermediate voices. Figure 2 compares the aver-
age LTAS of the students good, poor, and intermediate voices.
The poor voices differed from the good voices by having rela-
tively less sound energy at the range of 13 kHz. The good voi-
ces also had a more prominent peak at 34 kHz. The
intermediate voices did not differ from the good ones in terms
of the LTAS characteristics. The analyses were made using
FIGURE 1. Schematic picture of measurement of the level differ-
ence L1L0 between the F1 variation range (L1) and the F0 variation
range (L0) in LTAS. In this case the difference is +2.5 dB.
Timo Leino Students Voices 673

professional actors.33 Good voices were characterized by

FIGURE 2. Average LTAS for the male university students good
voices (N 14, dark line), poor voices (N 14, grey line), and inter-
mediate voices (N 22, dotted line).
a 10 kHz frequency range, but in figures the spectra are com-
pared in the range of 05 kHz as no significant differences
were found above 5 kHz.
Table 1 shows the statistical results. It can be seen that the
difference between the good and the poor voices was significant
in all other ranges studied except for 23 kHz and 45 kHz. The
difference between the good and the intermediate voices, in-
stead, was nonsignificant in all ranges. L1L0 level difference
did not distinguish the three voice quality groups either (good
voices: mean 0.64 dB, SD 3.6; poor voices: mean 2.11 dB,
SD 3.6 and intermediate voices: mean 0.27 dB, SD 4.2. Differ-
ences between the groups were nonsignificant).
Table 2 summarizes the results of Leq measurements. Stu-
dents with good voices had significantly higher Leq than those
with poor voices, while students with intermediate voices did
not differ significantly from those with good voices.
DISCUSSION
Differences in the LTAS of the voice quality groups among
students were basically similar to those reported earlier for
a less steep spectral slope and a more prominent peak at 3
4 kHz. This peak, however, was not as prominent in the good
students voices as in the best voices of the actors.33 For the stu-
dents the level of the 3.5 kHz peak was on the average 28 dB,
relative to the strongest peak in the LTAS. According to earlier
results, the level of this peak was on the average 20 dB for the
best voices of the actors. The correlation between the strength
of the peak and the perceptual voice quality was low in students,
whereas according to earlier results it was moderate in actors.33
Good voices did not differ from the ordinary voices among the
students. This acoustic finding is in accordance with voice
trainers perception that clearly good voices are quite rare
among vocally untrained university students. Among young
students this may be due to social pressure not to be different
from the others.
LTAS illustrates both phonatory and resonatory characteris-
tics. However, Nolan23 has found that LTAS is more capable
of distinguishing between phonatory than resonatory differ-
ences in voice qualities. The level differences between fre-
quency ranges in the LTAS reflect the slope of the source
spectrum, which in turn is related to the glottal closing speed.
Increased closing speed gives a less tilting slope.36 The spectral
tilt is known to decrease together with vocal intensity and effort
increase. This is due to the fact that the higher overtones gain
more in intensity than the lower ones when the overall intensity
is increased.37 In the present study the students with good and
ordinary voices used somewhat more intense voice than the
students with poor voices (mean difference 1.9 dB between
good and intermediate voices and 5.1 dB between good and
poor voices).
Sound level and voice quality seem to be interrelated. In gen-
eral, sound level tends to be somewhat higher after vocal exer-
cising or a longer voice training period.38,39 This may suggest

TABLE 1.
Differences between the average LTAS and Alpha ratio (level difference above and below 1 kHz): Leq
(15 kHz)Leq (50 Hz1 kHz) of good voices (N 14), poor voices (N 14), and intermediate voices (N 22)
of Finnish male university students (N 50 in total)
01 kHz 12 kHz 23 kHz 34 kHz 45 kHz Alpha
Good
Mean 7.6 dB 18.5 dB 27.5 dB 32.2 dB 42.4 dB 11.3
SD 1.8 3.3 4.0 4.1 2.8 (15.3 to 8.1)
Poor
Mean 9.8 dB 22.0 dB 30.0 dB 36.5 dB 44.5 dB 13.3
SD 2.1 3.5 2.5 4.1 5.0 (18.3 to 8.7)
Intermediate
Mean 8.3 dB 18.2 dB 26.5 dB 31.6 dB 42.3 dB 10.5
SD 2.1 4.3 5.7 6.0 3.8 (15.5 to 5.9)
Significance of differences
Good/poor P 0.006 P 0.012 NS P 0.015 NS P 0.043
Good/intermediate; NS NS NS NS NS NS
Mean and standard deviations are given for the averages of the normalized amplitude values in different frequency ranges. Statistical significance: MannWhit-
ney U test.
NS nonsignificant P > 0.05.
674
TABLE 2.
Differences between Leq of the students with good voices (N 14), poor voices (N 14),
and intermediate voices (N 22)
Good
Mean 72.8 dB
SD 2.7
Significance of differences
Good/poor P 0.001 Good/intermediate; NS
Statistical significance: MannWhitney U test.
NS nonsignificant P > 0.05.
that subglottic pressure is higher after training. We may ask
whether the only difference between speakers with good voices
and those with not so good voices is related to the habitual loud-
ness of speechand could thus be abolished by merely asking
the speakers with poor voices to speak more loudly. On the
other hand, somewhat higher sound level could also be a sign
of improved vocal function per se. For instance, there are results
suggesting that the glottal spectrum tilts less for trained singers
than for untrained subjects.38 This could result from a higher
glottal closing speed or a more favorable input impedance of
the vocal tract, for example, epilaryngeal narrowing.40 A faster
glottal closing speed could show a more adequate adduction
and/or improved mobility of the vocal folds. The latter could
possibly result from a more economic activity ratio between cri-
cothyroid and thyroarytenoid muscles.41 The strength of the 3
4 kHz peak compared to the strongest spectral peak (F0 or F1)
is naturally to some extent related to overall spectral slope, but
the prominence of that peak in the good voices of the actors
does not seem to be merely a result of a less tilting spectral
slope. According to the earlier studies by Leino,33 it was not
the level differences between frequency ranges but the promi-
nence of the 3.5 kHz peak in the average LTAS that distin-
guished the actors good voices from the fairly good and
rather poor voices. Below 3 kHz the relative sound level of
the rather poor voices was the same and that of the fairly
good voices was even higher than that of the good voices. In in-
dividual LTAS, it has been possible to see that the peak at
3.5 kHz, the actors formant, can be almost as strong or in
some cases even stronger than the preceding peak in the range
of 23 kHz.42 This suggests that the peak is not only a result of
strong voice source overtones in general but of a special reso-
nance type, most likely related to the actors technique of pro-
jecting. The region above 3 kHz is supposed to be more prone
to show resonatory than phonatory characteristics.24 The results
of Hurme43 also showed that when the speakers increased voice
intensity (556575 dB) in text reading (30 seconds), the level
differences between frequency ranges in the LTAS (strongest
peak set as zero) changed substantially more below 3 kHz
than above it. At 4 kHz the change was nonsignificant. This
was also demonstrated in Nordenberg and Sundberg.44
The clear valleys separating the 3.5 kHz peak from its sur-
roundings suggest that it is formed by a formant or a cluster
of formants. This peak may basically be related to the fourth
Journal of Voice, Vol. 23, No. 6, 2009
Intermediate Poor
70.9 dB 67.7 dB
3.6 4.0
Intermediate/poor; P 0.019
formant. According to the results of Iivonen and Laukkanen,45
the fourth formant for a male speaker of Finnish was located at
the range 3,3603,800 Hz depending on both the vowel and its
duration. Thus, the vowel-related changes in the frequency of
F4 were only minor. This suggests that F4 is not very heavily
dependent on articulation and thus most likely is not very lan-
guage dependent either. A fairly strong peak at 3.5 kHz can
also be seen in the LTAS of speakers of different languages: De-
jonckere published such an LTAS in a French speaking subject,6
Frkjr-Jensen and Prytz9 obviously from samples in Danish
and Nolan from a sample in English.23 Nawka found this
peak in good voices among German speakers.46 This peak,
however, is not an absolute prerequisite of a good voice quality,
and sometimes it can also be seen in not so good voices,47 for
example, in those with vocal fry.23
The level difference L1L0 has been found to correlate neg-
atively with the perception of breathiness (hypofunctional voice
production35). Furthermore, the level difference has been found
to rise after voice training while the perceptual voice quality
also became tighter (ie, more hyperfunctional).39 According
to the results of the present study, L1L0 difference was on av-
erage positive for the good and ordinary voices and negative for
the poor voices. This suggests that the relative level of the F0
was lower and voice production thus most likely tighter in the
students with good and ordinary voices compared to those
with poor voices. This may be related to greater loudness or oth-
erwise tighter adduction. Bele has also found a greater L1L0
and stronger sound level in actors compared to teachers.47 In the
present study, however, L1L0 difference did not differentiate
significantly between voice qualities of normal voiced un-
trained students. This may be due to the fact that not so good
voice may be either more hypofunctional or more hyperfunc-
tional than the optimum and thus at group level the differences
in this parameter will cancel each other out.
CONCLUSIONS
In vocally untrained male students speaking samples, those
which were perceived as representing a good voice quality dif-
fered from those perceived as representing a poor voice quality
by having a relatively higher sound level between 1 and 3 kHz
in the LTAS and a more prominent peak at 34 kHz. Good voi-
ces did not differ from intermediate voices. Leq was
Timo Leino Students Voices
significantly lower in samples evaluated as poor in voice qual-
ity. The level difference between the regions of the fundamental
and the first formant did not differentiate between voice quali-
ties.
The peak at 34 kHz (actors formant) cannot result from
louder voice alone but seems to imply some resonatory phe-
nomena as well. The effects of Leq and resonance characteris-
tics on the LTAS characteristics of trained and untrained good
voices need to be studied further.
Acknowledgments
The author is grateful to Mr. Jussi Helin for technical assistance
in the analyses and to Mrs. Virginia Mattila, M.A., for language
correction of the manuscript. Professors Johan Sundberg and
Anne-Maria Laukkanen are acknowledged for their valuable
comments.
REFERENCES
1. Ansi S. USA standard: Acoustical terminology (s 1.1). New York: Ameri-
can National Standards Institute, Inc.; 1960.
2. Titze IR. Principles of voice production. Englewood Cliffs, N.J: Prentice
Hall; 1994.
3. Li K-P, Hughes GW, House AS. Correlation characteristics and dimension-
ality of speech spectra. J Acoust Soc Am. 1969;46:1019-1025.
4. Majewski W, Rothman H, Hollien H. Acoustic comparisons of American
English and Polish. J of Phon. 1977;5:247-251.
5. Gauffin J, Sundberg J. Clinical applications of acoustic voice analysis
acoustical analysis, results and discussion. In: NH. Buch, ed. Proceedings
of the International Association of Logopedics and Phoniatrics Congress
1518 August 1977, Copenhagen, Denmark. Herning: Organizing Commit-
tee of the Congress, 1978:489502.
6. Formby C, Monsen RB. Long-term average speech spectra for normal and
hearing-impaired adolescents. J Acoust Soc Am. 1982;71:196-202.
7. Hurme P, Sonninen A. Normal and disordered voice quality: listening tests
and long-term spectrum analyses. In: Hurme P, ed, Papers in Speech Research,
Vol. 6. Jyvaskyla: Department of Communication, University of Jyvaskyla;
1985:49-72.
8. Dejonckere PH. Acoustic analysis of voice production. Essai de synthese
dans une optique clinique. Acta Oto-Rhino-Laryngologica Belgica. 1986;
40:377-385.
9. Frkjr-Jensen B, Prytz S. Registration of voice quality. Bruel Kjr Tech
Rev. 1976;3:3-17.
10. Prytz S. Long-time-average-spectra (LTAS) analyses of normal and patho-
logical voices. In: NH. Buch, ed. Proceedings of the International Associ-
ation of Logopedics and Phoniatrics Congress 1518 August 1977,
Copenhagen, Denmark. Herning: Organizing Committee of the Congress,
1978: 459475.
11. Wendler J, Doherty ET, Hollien H. Voice classification by means of long-
term speech spectra. Folia phon. 1980;32:51-60.
12. Hammarberg B, Fritzell B, Gauffin J, Sundberg J, Wedin L. Perceptual and
acoustic correlates of abnormal voice qualities. Acta otolaryngologica.
1980;90:441-451.
13. Dejonckere PH, Villarosa D. Long-term spectrum analysis of the voice.
Comparaison de voix normales et de voix alterees par differentes categories
de pathologies laryngees. Acta Oto-Rhino Laryngologica Belgica. 1986;40:
426-435.
14. Benson R, Hirsh I. Some variables in audio spectrometry. J Acoust Soc Am.
1953;25:499-505.
15. Pruzansky S. Pattern matching procedure for automatic talker recognition.
J Acoust Soc Am. 1963;35:354-358.
16. Niemller A, McCormick L, Miller J. On the spectrum of spoken English. J
Acoust Soc Am. 1974;55:461.
17. Kiukaanniemi H, Siponen P, Mattila P. Individual differences in the long-
term speech spectrum. Folia phoniatrica. 1982;34:21-28.
675
18. Harmegnies B, Landercy A. Intra-speaker variability of the long term speech
spectrum. Speech Commun. 1988;7:81-86.
19. Pavlovic V, Rossi M, Espesser R. Statistical distributions of speech of var-
ious languages. J Acoust Soc Am. 1990;88(Suppl 1):176.
20. Byrne D, Dillon H, Tran K, et al. An international comparison of long-term
average speech spectra. J Acoust Soc Am. 1994;96:2108-2120.
21. Williams CE, Stevens KN. Emotions and speech: some acoustical corre-
lates. J Acoust Soc Am. 1972;52:1238-1250.
22. Pittam J, Gallois C, Callan V. The long-term spectrum and perceived emo-
tion. Speech Commun. 1990;9:177-187.
23. Nolan F. The Phonetic Bases of Speaker Recognition. Cambridge, MA:
Cambridge University Press; 1983.
24. Kitzing P. LTAS criteria pertinent to the measurement of voice quality.
J Phon. 1986;14:477-482.
25. Pittam J. Discrimination of five voice qualities and prediction to perceptual
ratings. Phonetica. 1987;44:38-49.
26. Dmitriev L, Kiselev A. Relationship between the formant structure of dif-
ferent types of singing voices and the dimension of supraglottal cavities.
Folia phoniatr. 1979;31:238-241.
27. Rossing T, Sundberg J, Ternstrom S. Acoustic comparison of voice use in
solo and choir singing. J Acoust Soc Am. 1986;79:1975-1981.
28. Lofqvist A, Mandersson B. Long-time average spectrum of speech and
voice analysis. Folia Phoniatr. 1987;39:221-229.
29. Novak A, Dlouha O, Capoca B, Vohradnik M. Voice fatigue after theatre
performance in actors. Folia Phoniatr. 1991;43:74-78.
30. Rantala L, Paavola L, Korkko P, Vilkman E. Working-day effects on the
spectral characteristics of teaching voice. Folia Phoniatr Logop. 1998;50:
205-211.
31. Jonsdottir V, Laukkanen A-M, Siikki I. Changes in teachers voice quality
during a working day with and without electric sound amplification. Folia
Phoniatr Logop. 2003;55:267-280.
32. Leino T. The spectral characteristics of good voice. In: Finnish. Licentiate
Thesis in Logopedics. Helsinki: University of Helsinki, Department of Pho-
netics; 1976.
33. Leino T. Long-term average spectrum study on speaking voice quality in
male actors. In: Friberg A, Iwarsson J, Jansson E, Sundberg J, eds,
SMAC93, Proceedings of the Stockholm Music Acoustics Conference,
July 28August 1, 1993, No 79. Stockholm: The Royal Swedish Academy
of Music; 1994:206-210.
34. Sundberg J. Articulatory interpretation of the singing formant. J Acoust Soc
Am. 1974;55:838-844.
35. Laukkanen A-M, Vintturi J, Vilkman E, Sala E, Siikki I, Lukkarila P. Per-
ceptual, acoustic and self-reported correlates of vocal loading. In: Proceed-
ings of the XXVth world congress of the International Association of
Logopedics and Phoniatrics in Montreal.
36. Gauffin J, Sundberg J. Data on the glottal voice source behavior in vowel
production. In: Speech Transmission Laboratory, Quarterly Progress and
Status Report, 23. Stockholm: Royal Institute of Technology; 1980. 61
70.
37. Fant G. Speech Sounds and Features. Cambridge, MA: The MIT (Massa-
chusetts Institute of Technology) Press; 1973.
38. Wedin S, Leanderson R, Wedin L. Evaluation of voice training. In: N.H.
Buch, ed. Proceedings of the International Association of Logopedics
and Phoniatrics Congress 1518 August 1977, Copenhagen, Denmark.
Herning: Organizing Committee of the Congress, 1978:361381.
39. Laukkanen A-M, Syrja T, Laitala M, Leino T. Effects of two-month vocal
exercising with and without spectral biofeedback on student actors speak-
ing voice. Log Phon Vocol. 2004;29:66-76.
40. Titze IR, Story BH. Acoustic interactions of the voice source with the lower
vocal tract. J Acoust Soc Am. 1997;101(4):2234-2243.
41. Titze IR, Talkin DT. A theoretical study of the effects of various laryngeal
configurations on the acoustics of phonation. J Acoust Soc Am. 1979;66(1):
60-74.
42. Leino T, Karkkainen P. On the effects of vocal training on the speaking
voice quality of male student actors. In: Elenius K, Branderud P, eds,
Proceedings of the XIIIth International Congress of Phonetic Sciences,
Stockholm, Sweden 1319 August, 1995, Vol. 3 of 4. Stockholm: De-
partment of Speech Communication and Music Acoustics, Royal Insti-
676
tute of Technology and the Department of Linguistics, Stockholm Uni-
versity; 1995:496-499.
43. Hurme P. Acoustic Studies of Voice Variation. Doctoral thesis. In: Jyvaskyla
Studies in Communication, 7. Jyvaskyla: University of Jyvaskyla; 1996.
44. Nordenberg M, Sundberg J. Effect on LTAS of vocal loudness variation.
Logop Phoniatr Vocol. 2004;29:183-191.
45. Iivonen A, Laukkanen A-M. Explanations of the qualitative variation of
Finnish vowels. In: Iivonen A, Lehtihalmes M, eds. Studies in Logopedics
Journal of Voice, Vol. 23, No. 6, 2009
and Phonetics 4, Series B: Phonetics, Logopedics and Speech Communica-
tion, 5. Helsinki: Department of Phonetics, University of Helsinki; 1993:
29-54.
46. Nawka T, Anders LC, Cebulla M, Zurakowski D. The speakers formant in
male voices. J Voice. 1997;11(4):422-428.
47. Bele IV. Professional Speaking Voice: A Perceptual and Acoustic Study of
Male Actors and Teachers Voices. Doctoral dissertation. University of
Oslo; 2002.