Analysis of Acoustic Correlates of British ... - Brunel University

ANALYSIS OF ACOUSTIC CORRELATES OF BRITISH, AUSTRALIAN AND
AMERICAN ACCENTS
Qin Yan Saeed Vaseghi Dimitrios Rentzos *Ching-Hsiang Ho Emir Turajlic
Department of Electronic and Computer Engineering, Brunel University, Uxbridge, Middlesex, UK UB8 3PH
*Fortune Institute of Technology, Kaohsiung Taiwan, 842, R.O.C
(Qin.Yan, Saeed.Vaseghi, Dimitrios.Rentzos, Emir. Turajlic)@brunel.ac.uk *ch.ho@center.fjtc.edu.tw
ABSTRACT
This paper presents an analysis of the acoustic correlates
of the differences of British, Australian and American
English accents. The structures of the differences that
characterise accents in speech can be divided into two
parts: (a) phonetic differences and (b) acoustic
differences. The focus of this paper is on the analysis of
acoustic correlates of accents including formants and
their trajectories, pitch trajectory, pitch accent, pitch
nucleus, duration and speaking rate. The acoustics of
accents are modelled and estimated using twodimensional
HMMs of formants and a model of pitch
such as Rise/Fall/Connect (RFC) model. The differences
between British, Broad Australian and General American
English accents are discussed. Australian accent has a
lower 1 st formant (F1) but higher 2 nd formant (F2)
compared to British and American. The second formant
in speech is considered as the most sensitive to accent
identity. British speakers have the largest pitch frequency
range and the largest initial pitch rise and final pitch fall
rates in utterances. Australian accent exhibits significant
elongation of vowels and the lowest speaking rate
compared to other two accents. The differences in
acoustic correlates across accents are used to morph the
accent of a source speaker towards a target accent.
1. INTRODUCTION
Accent is one of the most fascinating aspects of speech
acoustics. An accent is a distinctive characteristic manner
of pronunciation, usually associated with a community of
people with a common regional or social/cultural
background. Accents are dynamic processes that evolve
over time influenced by large immigrations and social
and cultural trends. J.C. Wells [1] provides an excellent
introduction to the linguistic structures of accents of
English language. In [2,3] Watson and Harrington
describe a comparative analysis of formants of British
English, Australian English and New Zealand English
and three subclasses of Australian namely: Broad
Australian English, General Australian English and
Cultivated Australian English.
Accent is a relatively under-explored aspect of
automatic speech recognition (ASR) and text-to-speech
(TTS) systems. Accent variation is considered as one of
the dominant causes of deterioration in speech
recognition performance. In [4] Humphries improves
speech recognition accuracy by automatically modelling
pronunciations of the accents of the input speech. In [5]
Arslan and Hansen present an acoustic analysis of a
number of foreign English accents.
Accent Structure
Parameters
Comments
Phonetic Parameters
Substitution, Pronunciation differences from
insertion, deletion ‘standard’ phonetic transcription
Formant Correlates
Formants & their
trajectories
2 nd formant is most sensitive to the
accent
Intonation Correlates
F 0 range
Range of pitch
F 0 Utterance Slope Pitch slope across the utterance
Pitch Nucleus Prominent point (stressed) within an
intonation group (Tone Unit)
Initial Pitch Rising First pitch slope of a narrative
utterance
Final Pitch
Lowering
Final Pitch Rising
Final fall pitch slope of a narrative
utterance
Final rise pitch slope of a narrative
utterance
Duration and Speaking Rate Correlates
Speaking Rate Phonemes or words per second
Phoneme Duration Vowel duration elongation and
complete pronunciation all affect
duration
Table 1: A list of phonetic and acoustic correlates of accent.

Speakerindependent
speech database
(UK,AU,US)
Feature
extraction
Pitch trajectory
estimator
HMM
training
Forced
alignment
segmentation
Pitch pattern
analysis
This paper presents an outline of an accent model in
terms of the linguistics and acoustics correlates of an
accent. Experiments are focused on three major accents
of the English language namely British, Broad Australian
and American. The databases used are the Australian
accent speech ANDOSL database, American accent
speech WSJ database (Table 2) and British accent speech
WSJCAM0. The features used to train hidden Markov
models of speech are 39 Mel-Frequency Cepstral
Coefficient (MFCCs) with energy and their
differentiation and acceleration. For segmentation and
labeling of speech left-to-right HMMs of triphone units
with 3 states and 20 Gaussian mixtures per state are used.
2. OVERVIEW OF ACCENT
MODELLING
Formant
2D HMMs
Duration &
speaking rate
models
Pitch trajectory
model
Figure 1 An Overview of Acoustic Modelling of Accents.
The structural differences of accents can be divided into
two broad parts: (a) differences in phonetic transcription
and (b) differences in acoustic correlates. Table 1 shows
a list of the acoustic features that represent the acoustic
space of an accent with a brief comment on each set of
correlates. Figure 1 illustrates the feature extraction and
modelling process involved in quantifying the correlates
of an accent in terms of the values of the formants, pitch
and duration. Formants of each phoneme are modeled by
2D HMM [6,7]. A 2D HMM is modeled as a
combination of a 1-D HMM along time dimension and a
1-D HMM along frequency dimension. Along the
frequency axis, every state of a 2D-HMM of formant
models the distribution of one formant of the phoneme.
Pitch trajectories are used to model the broad intonation
characteristics pattern of an accent. Duration and
speaking rate are modelled by traditional speech HMMs.
3. PHONETIC CORRELATES OF ACCENTS
The most dominant aspect of accents is often manifested
Database Name
(accent)
No.
Speakers(f/m)
No.
Sentences
ANDOSL (Australian) 18/18 7200
WSJ (American) 36/38 9438
WSJCAM0(British) 40/46 9476
Table 2: Databases configurations
Word American British Australian
John ʤΛn ʤƆn ʤƆn
day dei dei dæi
immediate ı’mi:ʤ^t i’mi:ʤət ə’mi:di:ət
chassis ʃæsı ʃæ sı ʃæzi:
Table 3: Examples of Difference in Phonetic Realization.
by the differences in pronunciation as transcribed by the
phonetic dictionary. These phonetic transcription
differences can be broadly categorised into two classes:
(a) Differences in the phonemic alphabet systems,
i.e. in the number or identity of phonemes. For
example, for automatic speech processing,
British accent as transcribed by Cambridge
University’s BEEP dictionary has five extra
vowels: /ax ea ia ua oh/ compared to American
as transcribed by Carnegie Melon University’s
CMU dictionary. Australian English has
distinctive vowels such as /æi/ instead of /ai/
and /æu/ for /au/.
(b) Differences in phonetic realisation including
phoneme substitution / deletion / insertion.
Table 3 gives some examples of phonetic
differences between American, Australian and
British accents.
4. ACOUSTIC CORRELATES OF ACCENTS
4.1 Formant Correlates of Accent
Formants are the main sources of acoustic ‘colouring’ of
F1 (Hz)
F2 (Hz)
1000
900
800
700
600
500
400
300
200
2800
2600
2400
2200
2000
1800
1600
1400
1200
1000
800
600
Australian British American
AA AE AH AO EH ER IH IY OH UW UH
Australian British American
AA AE AH AO EH ER IH IY OH UW UH
Figure 2: Formants Comparison of Australian, British and
American (female).

U 1
U 2
U 3
T 1 T 2 T 3 T 4 T 5
E 1 E 2 E 3
A 1 A 2 A 3 A 4 A 5
Figure 3: Formant Space of Australian, British and
American (female).
speech. Part of the acoustics of an accent is due to the
differences in the distribution of formant frequencies of
vowels and diphthongs. In [2,3] the differences between
the distributions of the formants of the vowels of
Australian and New Zealand English are shown. In this
paper experiments are conducted on an enlarged accent
set.
Formant frequency estimation is based on an
improved LPC-based method and 2-D HMM [6,7]. The
average first and second formants of American, British
and Australian English are shown in Figure 2. It can be
seen that across the accents the second formants of
vowels have the largest frequency range of up to 2KHz
and also some of the highest frequency differences of
formants. It is interesting to note that the second formant
which is most affected by accent also coincides with a
Figure 4: Average Formant Trajectories (female) of British,
Australian and American after alignment.
X-axis is the normalized time. Y-axis is frequency (Hz).
relatively high sensitivity frequency band of hearing.
Figure 5: The hierarchy structure of the prosodic across the
pitch contour. The unit of x-axis is 10ms and the unit of y-
axis is Hz. ‘T’ denotes the tone units. ‘E’ denotes the pitch
event. ‘A’ denotes the pitch accent [6].
The 2 nd formant of vowels in Australian are on
average 11% higher than those of British and 8% higher
than those of American. The first formant of vowels
varies within a 1KHz frequency range. Experimental
results show that Australian vowels have a lower first
formant compared to those in British and American. The
third and fourth formants display a difference of some
10% across accents. Figure 3 illustrates the formant
space of three major English accents [7]. Some features
noted from Figure 3 are:
Rising of vowels /ae/ and /eh/ in Australian.
Fronting of the open vowel /aa/ and the high
vowels /uw/ and /uh/ in Australian.
Fronting and rising of the vowel /er/ in
Australian.
The vowels /iy/, /eh/ and /ae/ in Australian are
closer. It is particularly noticeable that for some
vowels such as /ih/ /uw/ /ae/ /eh/, their positions
in F1/F2 panel are shifted significantly with
accents.
Figure 4 shows the time-normalised average formant
trajectory of British, Australian and American English
accents. The acoustic target point of the formant
trajectory of each vowel is marked as the point where the
formant reaches the steadiest value (“a static spectral
slice [8]”). It is noticeable in Figure 4 that Australian
vowels have delayed target points compared to British
and American vowels.
4.2 Pitch Intonation Correlates of Accent
Much of an accent is conveyed by intonation, which in
turn is largely conveyed by pitch trajectory. Pitch
trajectory is hence regarded as one of the most important

correlates of accents. In [9] Cruttenden points out that the
initial rise or the final fall/rise in pitch is an indicator of
different accent types. In particular, Scottish and
Northern Ireland English tend to use final pitch rise in
declarative utterance.
The Rise/Fall/Connect (RFC) model [6,10] is
employed to describe the pitch contour differences in
accents modelled via Legendre polynomial function.
The pitch correlate of intonation has the following
hierarchical structure [6] (Figure 5)
1) Pitch Accent: this is either a pitch rise or a fall.
2) Pitch Event: a combination of a pitch rise and
pitch fall (or vice versa)
3) Tone Unit: a continuous sequence of several
pitch events within phrase or utterance.
4) Utterance: the composite trajectory of the pitch
signal across a number of tone-units in an
utterance.
Tone unit must contain at least one prominent pitch
event. This prominent pitch event is called the pitch
nucleus. It carries the major part of intonation
information. The magnitude and relative position of the
Gender Female
Male
Accent (Avg.Pitch 3*std) (Avg.Pitch 3*std)
British 184 3*37.30 121 3*29.93
Australian 194 3*21.21 119 3*17.98
American 195 3*21.78 121 3*17.17
Table 4: Average Pitch and Pitch Frequency Range (Hz) of
British, Australian and American accents.
Gender
Accent Female Male
British 13.37 8.53
Australian 14.77 8.89
American 14.91 9.07
Table 5: Pitch Utterance Slope (Hz/sec).
Gender
Accent Female Male
British 74.77 60.94
Australian 44.36 31.6
American 58.72 46.31
Table 6: Initial rise pitch rate (Hz/sec).
Gender
Accent Female Male
British 244 146
Australian 68 88
American 225 116
Table 7: Final lowering pitch rate (Hz/sec).
Frequency
Initial
Pitch
Rise
Pitch Range
Pitch Sentence Slope
Pitch nucl eus
Final Pitch Fall
This warning / now repr e sents / a m i nority view
Figure 6: Broad Pitch Intonation Pattern in Accent Modelling
Thick line is the pitch contour.
Time
nucleus is correlated with accents. Different accents tend
to put nucleus in different positions [1].
For accent modelling, in this paper only the broad
intonation patterns are modelled (Figure 6). A set of
parameters are considered including pitch range, the rate
of initial pitch rise in an intonation utterance, the rate of
final pitch fall and pitch slope across an utterance.
Table 4 gives the estimates of average pitch frequency
and pitch range of male and female speakers across
British, Australian and American accents. There is no
significant difference in average value of F 0 across the
three accents. This implies that these accents do not have
a significant impact on average pitch frequency. Hence
average pitch frequency can be considered as a more
speaker-dependent parameter. Table 4 also shows that
female speakers have a larger pitch frequency range than
male speakers. Both male and female British speakers
have the largest pitch frequency range among the three
accents. On average the pitch frequency range of British
speakers is 41% and 43% larger than those of Australian
and American respectively. Australian and American
speakers possess similar pitch frequency range. It is
believed that pitch frequency range is closely related to
intonation patterns employed by different accents.
British speakers have extensive usage of low-rise tone in
non-final intonation group (tone unit) to form an
Duration (s)
0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
Australian British American
aa ae ah ao aw ay eh er ey ih iy ow oy uh uw
Figure 7: Comparison of duration of Australian, British and
American vowels.

Duration (s)
0.18
0.15
0.12
0.09
0.06
0.03
0
Australian
British
American
b ch d dh f g hh jh k l m n ng p r s sh t th v w y z zh
Figure 8 Comparison of duration of British, Australian, American consonants.
oratorical and formal speaking style whereas Americans
prefer the high-rise tone, a more causal way. There is
also increasing tendency of using high-rise tone in
Australians [9].
Table 5 presents the pitch utterance slope. Female
speakers consistently have larger pitch contour slope than
male speakers regardless of their accents. It can be noted
that American has a slightly sharper slope than British
and Australian, with a difference of 8% and 3%
respectively.
Table 6 and 7 display the initial pitch rise rates of
British, American and Australian accents. It is noticeable
that female speakers consistently have steeper pitch rate
in the Initial rising part of pitch contour than those of
male speakers. British speakers possess the steepest rate
of initial pitch rise among the three accents about 44.1%
and 22.7% steeper than Australian and American
respectively. This coincides with the fact that they have
the largest frequency range of three. Similarly, British
speakers possess the sharpest final fall rate in pitch
compared to Australian and American speakers. Results
illustrate that British speakers tend to have a steeper pitch
rise and fall rates than American speakers. Furthermore,
American speakers tend to have a lower pitch in final
words of sentences compared to British speakers [11].
4.3 Phoneme Duration Pattern and Speaking Rate of
Accent
Accents are also partly conveyed by vowel duration and
speaking rate. Duration statistics is estimated using
segmentation boundaries obtained by HTK. Figure 7 and
8 displays the average vowel and consonant duration of
three accents. On average, the duration of Australian
vowels is 27% and 25% longer than those of British and
American accents respectively. In particular, diphthongs
such as /ao/ and /ow/ in Australian are over 33% longer
than those in British, and over 30% longer than those in
American. British speak shorter vowels at the starts and
the ends of sentences in comparison to American
speakers [11]. This could be due to British speakers
always tending to pronounce last syllable fast. As for the
consonants, Australian consonants are 5% and 6% longer
than American and British respectively. Duration of
nasal consonants /m n ng/, unvoiced stops /p t k/ and
semivowels /w l r/ in American are longer than those of
British. Affricates consonants /jh ch / in American are
shorter than those in British. Australian has longer /dh
hh/, shorter /z/ compared to British and American.
Speaking rate describes the speaking speed and can
be indicated by the number of syllables or words spoken
per minute and the duration of pause. In our analysis, it is
found that average speaking rate varies across accents.
Table 8 and Figure 9 present the figures and distributions
of the average speaking rates of British, Australian and
Figure 9: Distribution of speaking rate in American, British
and Australian accents.
Speaking Rate
(number/sec) Phone Word
British 12.1 3.64
American 11.6 3.1
Australian 10.8 2.8
Table 8: Speaking rates of British, American and Australian.
Model
Input
British
model
American
model
Australian
Model
British 12.8 29.3 34.9
American 30.6 8.8 29.94
Australian 33.1 27.3 7.28
Table 9: Word error (%) of cross accents speech recognition
between British, American and Australian.

American accents. Speaking rates of Australian English
and American English are respectively 23% and 15%
lower than that of British English. The fact that
Australian English possess the lowest speaking rate also
conforms to the elongation of durations of Australian
vowels. Unlike pitch rate, female speakers of three
accents do not illustrate consistently higher speaking
rates than male speakers.
It is also interesting to notice the apparent correlation
between the speaking rate and the accuracy of automatic
speech recognition. American and Australian English
respectively achieves 31% and 43% less word
recognition error than the British English in matched
accent conditions. Australian with the slowest average
speaking rate gives the best speech recognition rate of the
three accents as shown in Table 9. The reason may be
that the slower speaking rate leads to a more consistent
pronunciation and a better recognition results.
4. ACCENT CONVERSION
The differences of acoustic correlates across accents are
used to morph the accents of spoken sentences among
British, Australian and American accent speech
databases. The changes in the acoustic correlates of
accents of speech are accomplished by the VoiceMorph
software developed in our laboratory. VoiceMorph can
independently change all acoustic correlates of a voice.
Initial experiments were conducted by mapping the
phonetic, formants, pitch and duration correlates of
accents of speech between British, Australian and
American English accents. The results show that acoustic
correlates are an effective description of an accent for
changing the perceived accent of a voice from a source
accent to a target accent.
5. CONCLUSION
This paper presents a comparative analysis of acoustic
correlates of accents. Differences in accent are classified
into two categories: phonetic transcriptions and acoustic
correlates. Acoustic correlates of accents such as
formants are analysed for three major English accents.
These acoustic correlates are proved very effective in
accent morphing. However, the applications of the
analysis also include accent recognition and
compensation in automatic speech recognition, which are
subjects of further investigations.
7. REFERENCE
[1] Wells J.C., Accents of English, volume 1,2,3,
Cambridge University Press (1982).
[2] Harrington J., Cox F., Evans Z., “An Acoustic
Phonetic Study of Broad, General, and Cultivated
Australian English Vowels”, Australian Journal of
Linguistics 17: 155-184 (1997)
[3] Watson C. I., Harrington J., Evans Z., “An Acoustic
Comparison between New Zealand and Australian
English Vowels”, Australian Journal of Linguistics
(1996)
[4] Humphries J., “Accent Modelling and Adaptation in
Automatic Speech recognition”, PhD Thesis, Cambridge
University (1997)
[5] Arslan L., Hansen J., “A Study of Temporal Features
and Frequency Characteristics in American English
Foreign Accent”, Journal of Acoustic Society of America,
vol. 102(1), p. 28-40, (1997)
[6] Ching-Hsing Ho, “Speaker Modelling for Voice
Conversion”, PHD thesis, Department of Electronic and
Computer Engineering, Brunel University (2001)
[7]Qin.Yan, Saeed Vaseghi “Analysis, Modelling and
Synthesis of formants of British, American and
Australian Accents” ICASSP, Hong Kong, p. 712-715
(2003)
[8] Harrington, J., Cassidy, S. “Dynamic and Target
theories of Vowel Classification: Evidence from
Monophthongs and Diphthongs in Australian English”
Language and Speech 37 pp.357-373 (1994)
[9] Cruttenden A., Intonation, Cambridge University
Press (1997)
[10] Paul Taylor, “Analysis and Synthesis of Intonation
using Tilt Model”, Journal of the Acoustical Society of
America Vol 107 3, pp.1697-1714 (1994)
[11] Qin Yan, Saeed Vaseghi, “A Comparative Analysis
of UK and US English Accents In Recognition and
Synthesis”, ICASSP (2002)
6. ACKNOWLEDGEMENTS
We would like to thank the UK’s EPSRC for funding
project no GR/M98036.