Overview on age markers in the voices of adult healthy speakers In the past two or three decades, many studies were concerned with the features of speakers' age. We are concentrating on the most stable features in adult healthy speakers, excluding voice

1
Age-related changes in fundamental frequency and
formants a longitudinal study of four speakers.
Jonathan Harrington1, Sallyanne Palethorpe 2,
Catherine I. Watson3
1. Institute of Phonetics and Speech Processing
(IPS), University of Munich, Munich, Germany. 2
Macquarie Centre for Cognitive Science (MACCS),
Macquarie University, Sydney, Australia. 3 Dept.
of Electrical and Computing Engineering, The
University of Auckland, New Zealand.
2
Motivation for the present study sound change
and age
Vowel quality in adults even those who live all
their lives in the same speech community
changes in time.
Some changes are due to phonetic sound changes in
the community e.g. an analysis (Harrington et
al, 2000, 2007) of the Queen's Christmas
Broadcasts shows a fronter /u/ (GOOSE) and lower
/a/ (TRAP) which have both been shown to be sound
changes taking place in the standard accent of
English.
1.
But some vowel quality changes in adulthood are
caused by physiological processes of vocal tract
maturation and there is certainly extensive
evidence that age can be estimated perceptually
(e.g., Jacques Rastatter, 1990 Huntley,
Hollien, Shipp, 1987 Neiman Applegate, 1990
Ptacek Sander,1966)
2.
Therefore in order to understand 1. we must be
able to normalise for 2. (see also Price 2006,
Aus. Speech Science Technology Conference, for
a similar analysis/arguments w.r.t. Australian
English)
3
Motivation longitudinal study
Verdonck-de Leeuw and Mahieu, 2004, Journal of
Voice 'Longitudinal studies on vocal aging are
scarce..'
4
Age and f0
1. Many studies show a correlation between
increasing age and decreasing fundamental
frequency (Helfrich, 1979 Decoster Debruyne,
2000Linville, 2001 Linville Fisher, 1985)
although there is disagreement on whether this
applies equally to men and women.
2. Physiological interpretation (e.g. Linville,
2004) "10-15 Hz drop in women at menopause
resulting from hormonal changes that cause
thickening of the laryngeal mucosa. In men, f0
lowers approximately 10 Hz from young adulthood
to middle ageAfter middle age, f0 in men rises
substantially (approximately 35 Hz) into advanced
old age, reaching the highest level of adulthood."
5
Age and formants
Formant lowering with increasing age Linville
Fisher, 1985, Xue Hao 2003 Watson Munson,
2007
6
Age and formants
On the other hand, with increasing age
Xue Hao (2003) report only oral but not
pharyngeal tract lengthening and changes
predominantly to F1.
Meurer et al (2004) less variation but no
formant change
Rastatter et al. (1997) found consistent lowering
only in F1 and different changes in men and
women.
Age-related formant changes are vowel specific
(Rastatter et al, 1997 Watson Munson, 2007)
Labovs (1994) analysis of the same person aged
68 and then 85 no evidence of any significant
long-term change to F1 or F2 over these years.
7
Method Speakers
Two speakers data from several years over
roughly a 50 year period
Four speakers data from two years 30-40 years
apart.
8
Method
We report results on mean f0 and mean F1-F3 (a)
in schwa vowels (b) across all periodically
voiced frames
Since we obtained a very similar pattern of
results from (a) and (b), we present only the
results from (b) here.
9
Results I Mean f0 and mean F1 in 6 speakers.
Mean f0 and F1 from early (in the 1940s/1950s vs.
late (1980s/1990s) broadcasts in all 6 speakers.
mean f0
mean F1
early
500
200
late
300
Frequency (Hz)
100
50
100
0
0
(Speakers from L to R Cooke, Plomley, Bradman,
Hillary, Lockwood, Queen)
f0 is significantly lower in later broadcasts
F1 is significantly lower in later broadcasts.
10
Results II F2 and F3
Mean F2 and F3 from early (in the 1940s/1950s vs.
late (1980s/1990s) broadcasts in all 6 speakers.
F3
F2
1.8
2.8
early
1.6
2.6
Frequency (kHz)
1.4
2.4
late
1.2
2.2
C
P
B
H
L
Q
C
P
B
H
L
Q
(Speakers from L to R Cooke, Plomley, Bradman,
Hillary, Lockwood, Queen)
No consistent or significant effect across the 6
speakers on average F2 nor average F3 from early
to later broadcasts
11
Results III f0 and F1 averaged separately by
year over a 50 year period in the Queen and
Alistair Cooke
Hz
Hz
f0
F1
550
550
280
280
Queen
500
500
Frequency
240
240
450
450
200
200
400
400
Hz
Hz
130
130
400
400
Cooke
120
120
Frequency
360
360
110
110
320
320
100
100
Decade
12
Do f0 and F1 decrease at a similar rate?
If so, then either r in the exponential
regressions for f0 and F1
or equivalently, the slope, log(r), of the
straight line in the corresponding log domain
should be the same.
13
Do f0 and F1 decrease at a similar rate?
Queen
Cooke
F1
Log frequency
f0
Decade
F-tests showed that there is no significant
difference between the slopes for F1 and f0
(between the red and black line slopes), in the
log frequency domain, neither for the Queen, nor
for Cooke.
14
Results IV F2 and F3 averaged separately by year
over a 50 year period in the Queen and Alistair
Cooke
F2
F3
2.9
1.8
Queen
2.8
1.7
Frequency (kHz)
50
70
90
50
70
90
2.60
Cooke
1.50
2.55
1.46
2.50
Decade
50
60
70
80
90
50
60
70
80
90
No significant linear or exponential trend in F2
or F3 as a function of year for either the Queen
or Cooke.
15
Summary of results
(a) across two years
(b)
approx. every other year over a 50 year period in
two speakers (Cooke, Queen)
No consistent effects in F2 nor F3 in either (a)
or (b)
16
Interpretations age and vocal/oral tract
lengthening
It seems unlikely that the vocal tract lengthens
with increasing age, given the lack of
consistent effects in F2 and F3.
Also, when in Cooke's (much) later years, the
trend is reversed and f0 begins to rise, then so
does F1
The falling (age 40-81) then rising (age 82-94)
trend in F1 is not likely to be explicable by
vocal tract lengthening then shortening.
17
So we probably instead need to look for
physiological or auditory mechanisms that affect
f0 and F1 in equal measure with increasing age.
18
Auditory explanation
The difference f0 (Bark) F1 (Bark) has been
shown in various studies by Traunmüller to be
related to phonetic vowel height. Perhaps
speakers compensate for a physiologically induced
f0-lowering by lowering F1 (so that phonetic
vowel height stays roughly the same).
19
Physiological explanation
1. Various studies point to a reduction in the
pulmonary function (Ptacek et al, 1966),
laryngeal cartilage ossification, an increased
stiffening of the vocal folds (Kahane, 1987), and
a reduction in vocal fold closure (Ferrand, 2002)
with increasing age.
2. This suggests that it may become more
difficult to develop the same acoustic power for
the same degree of physiological effort with
increasing age.
3. Traunmüller and Ericksson (2000) have found
that greater vocal effort produces a boost to f0
and F1. Moreoever, when vocal effort increases
they found, (just like in this study) that
log(f0) covaries almost linearly with log(F1).
20
Thank you!