The ANDOSL Project commenced formally with the support of the Australian Research Council for a significant data collection of the spoken language of Australia in 1992. This had been preceded by some active planning by members of the Australian Speech Science and Technology Association (ASSTA) resulting in a number of papers dating from 1989, the establishment of a National Spoken Language Database (NSLD) committee in November 1990, a National Workshop in November 1991, and some preliminary funding to ANU and Macquarie University for compatible hardware in 1991.
The NSLD committee of ASSTA which initiated the ANDOSL project comprised Bruce Millar (ANU -chair), Phillip Dermody (National Acoustic Labs), Jonathan Harrington (Macquarie University), and Julie Vonwiller (Sydney University).
During 1993-1995 a significant programme of data collection was performed
and the products of this work are being made available progressively.
ANDOSL Project History
The Australian National Database Of Spoken Language (ANDOSL) is named
after the project which created the first core component of the ANDOSL
cluster of spoken language corpora and operated in the period 1993-95.
The concept of a National Speech Database in the Australian context
was presented to a European Speech Communication Association (ESCA)
Tutorial and Research Workshop in the Netherlands (Millar, 1989).
Millar reviewed the various Australian efforts of the 1980s to create
spoken language data corpora, and pointed to the role of a national
spoken language resource within the speech processing community. A
major emphasis of this paper was the need for a tightly-coupled
internal organisation of such a resource so that it would be both
accessible and intelligible.
Following the ESCA Workshop, Millar convened a small group of the
members of the Australian Speech Science and Technology Association
(ASSTA) in Sydney to assess support for this concept. A sub-group
of
four people Phillip Dermody (National Acoustic Laboratories), Jonathan
Harrington (Speech Hearing and Language Research Centre, Macquarie
University), Bruce Millar (Computer Sciences Laboratory, Australian
National University) and Julie Vonwiller (Speech Group, Electrical
Engineering Department, Sydney University) agreed to initiate some
actions towards achieving the goal of a National Speech Database.
The
aims of this group were endorsed by the executive of ASSTA on 16th
November 1990 in the formal terms of reference of the ASSTA National
Spoken Language Database (NSLD) committee, which comprised the same
four people.
In the same month that the ASSTA NSLD committee was formed two papers
were presented, one to the inaugural ICSLP conference in Kobe, Japan
(Millar et al, 1990a), and another to the third SST conference in
Melbourne, Australia (Millar et al, 1990b). These papers developed
further the concepts proposed in 1989 and exposed them to wider
scrutiny.
In 1991 Harrington and Millar were successful in obtaining Australian
Research Council (ARC) funding for a collaborative research infrastructure
project to develop appropriate hardware and software environments to
support cross-institutional collaboration in speech data collection,
annotation, and management. During the same year, Millar developed
a
descriptive syntax for spoken language data and circulated this widely
via electronic mail, seminars and conferences overseas and in Australia
(Millar, 1991a; 1991b). A National Workshop was held in Melbourne
in
November 1991 which brought together a wide range of speech researchers
from Engineering, Linguistics, and Clinical backgrounds to consider
the
establishment of a national corpus. A grant from Macquarie University
brought Dr Jim Hieronymus from Edinburgh University to give a keynote
talk
at that workshop.
In 1992 the NSLD committee resolved to apply for a large ARC research
infrastructure grant to launch a major core component of the desired
national database. It was at this stage that the name ANDOSL
was born.
Diverse activities preceded the application for the grant again
producing two conference papers (Croot et al, 1992; Millar, 1992).
An initial grant of $250,000 was awarded for 1993 which was an intense
year of detailed design work, staffing the project, and getting started
with data collection. A further extension grant of $150,000 was
awarded
for 1994 and the data collection was completed in that year.
A variety
of developments directly and indirectly related to the developing ANDOSL
corpus were reported (Harrington et al., 1993; Millar, 1993; Millar
et
al., 1994; Cassidy et al., 1994; Millar & Davies, 1994; Clerigh
&
Vonwiller, 1994).
The tasks of 1995 included consolidation of all the data into a unified,
cohesive, and structured corpus, much continuing work on labelling
at
both word and phonemic levels and including the development and testing
of automatic techniques, and general quality checking of all forms
of
data. The task of designing CDROMs and collating a diverse range
of
annotation data for dissemination was also completed during this year.
Some funds were carried over into 1996 to enable the completion of a
set
of master CDROMs covering all divisions of the data collected. Fifteen
CDROMs covering read sentence data, both from the Australia-born and
overseas-born speaker sets, and read isolated words from all speakers
were released in July 1996. Work on a further fifteen CDROMs
containing
the spontaneous speech MAP task data from all speakers has proceeded
in
the latter half of 1996 and their release is expected in early 1997.
The work of this truly collaborative project will continue to be
described in conference and journal presentations (eg Millar, 1995;
Vonwiller et al., 1996) but the most significant indication of the
success of this work will be the list of research publications and
speech technology developments which cite the use of this data in the
intellectual and commercial life of Australian speech scientists.
The aim of ASSTA's NSLD committee has been achieved with respect to
a
core segment of spoken language in Australia and that is very satisfying.
There is however much more that can be done building on the foundation
that has been achieved. It is intended that the dissemination
of the
existing core data will proceed under an appropriate management scheme
and that ANDOSL will be extended beyond the core material as the NSLD
committee continues to stimulate interest within the research community
to conceive, design, plan, and collect data corpora which are compatible
with this initial set of data.
Bruce Millar - 12 November 1996
Millar,J.B. (1989) Design and use of a national speech database,
In "Proceedings of the ESCA workshop on Speech Input/Output Assessment
and Speech Databases", Noordwijkerhout, 20-23 September, pp.2.5.1-2.5.4.
Millar,J.B., Dermody,P., Harrington,J.M., Vonwiller,J. (1990a)
A national spoken language database: concept, design, and implementation,
In "Proceedings of International Conference on Spoken Language Processing
(ICSLP-90)", Kobe, Japan, 18-22 November, pp.1281-1284.
Millar,J.B., Dermody,P., Harrington,J.M., Vonwiller,J. (1990b)
A national cluster of spoken language databases for Australia,
In "Proceedings of Third International Australian Conference on Speech
Science and Technology", Melbourne, 27-29 November, pp.440-445.
Millar,J.B. (1991a) Knowledge of speaker characteristics: its benefits
and
quantitative description, In "Proceedings of 12th Int Congress of
Phonetic Sciences", Aix-en-Provence, 19-24 August, pp.538-541.
Millar,J.B. (1991b) Towards a standard description of spoken language,
In "Proceedings of Workshop on International Cooperation and Standardisation
of Speech databases and speech I/O Assessment Methods", Chiavari, Italy,
26-28 September (no page numbers).
Croot,K., Fletcher,J., Harrington,J. (1992) Levels of segmentation and
labelling in the Australian National Database of Spoken Language,
In "Proceedings of the Fourth Australian International Conference on
Speech
Science and Technology" Brisbane, Australia, 1-3 December, pp86-90.
Millar,J.B. (1992) The description of spoken language,
In "Proceedings of the Fourth Australian International Conference on
Speech
Science and Technology" Brisbane, Australia, 1-3 December, pp.80-85.
Harrington,J.M., Cassidy,S., Fletcher,J., McVeigh,A. (1993)
"The mu+ system for corpus based speech research",
Computer Speech and Language Vol.7, pp305-331.
Millar,J.B. (1993) "The ANDOSL Project and its data description",
In "Report on the COCOSDA Workshop", Berlin, pp17-18.
Millar,J.B.,Vonwiller,J.P.,Harrington,J.M.,Dermody,P.J. (1994)
"The Australian National Database Of Spoken Language",
In "Proceedings of ICASSP-94", Adelaide, Vol.1, pp.97-100.
Cassidy,S., Harrington,J.M.. McVeigh,A. (1994)
"The mu+ Speech Database System",
Technical Report, SHLRC, Macquarie University
Millar,J.B., Davies,D. (1994)
"The ANDOR interface to the Australian National Database of Spoken
Language",
In "Proceedings of Fifth Australian International Conference on Speech
Science and Technology", Perth, Australia, 6-8 December 1994, pp.209-214.
Cleirigh,C., Vonwiller, J. (1994)
"Accent Identification with a view to Assisting Recognition",
Proceedings of ICSLP '94, pp.375-378
Millar,J.B. (1995)
"Overview of Australian Spoken Language Corpus resources and Standards",
In "Notes from the COCOSDA Workshop 1995".
Vonwiller,J., Rogers,I., Cleirigh,C., Lewis,W. (1996)
"Speaker and Material selection for the Australian National Database
of Spoken Language",
Journal of Quantitative Linguistics Vol.2:3, pp.177-211.
Vonwiller,J., Cleirigh,C., Garsden,H., Kumpf,K., Mountstephens,R., Rogers,I.
(1997)
"Development and application of an accurate and flexible automatic
aligner",
International,Journal of Speech Technology, Vol.1, pp.151-160.
Millar,J.B., Harrington,J.M., Vonwiller,J.P. (1997)
"Spoken Language Resources for Australian Speech Technology",
Journal of Electrical and Electronic Engineering Australia, Vol.17,
No.1, pp.13-23.