Munich
Thursday 25th and Friday 26th May, 1995
We are currently starting the preparations for a two-day
workshop on articulatory databases. This will be the third in
the series of workshops organized by the ACCOR working group
(a consortium of phonetic institutes financed by the European
Community's ESPRIT programme) and and follows the
Electromagnetic Articulography meeting in Munich (April,
1992), and the Tongue Modelling meeting in Barcelona (December
1993).
A few words on the aims of a meeting devoted to such an
apparently dry topic:
The basic premise is that the free availability of
articulatory data could provide benefits in several partly
overlapping areas:
In basic research it could allow investigators to test
hypotheses formulated in articulatory terms on a much
wider range of data than the individual worker would
normally be able to acquire or access unaided.
It could promote the development and testing of
algorithms for deriving articulatory representations from
acoustic data - relevant both for basic understanding of
speech production as well as in potential applications
such as speech displays for training the speech impaired.
It could promote the development and testing of
algorithms for speech synthesis/recognition using an
articulatory level of representation.
The aim of the workshop would be to generate an exchange of
ideas among people active in these areas in order to identify,
for example (further suggestions welcome):
1) What articulatory data is in existence that it would be
beneficial to make more freely available
(archival/retrospective approach)?
2) What standards should freely available data meet?
- Specification of recording conditions
- Anatomical frames of reference
- Levels of accuracy/reliability
- Linguistic specification of the speech samples
- Preferred data structures for distribution.
- Any other issues relevant to the data being used
without risk of misinterpretation by people not actually
involved in the the details of acquisition
Articulatory data can come in many guises depending on
the speech subsystem tapped into and the transduction
technique used. Thus, it is possible to monitor position
(1, 2 or 3 dimensions), force, EMG, airpressure/flow with
techniques that may be static or dynamic, may involve
imaging or point-tracking etc. etc. We think that people
directly involved in acquisition can also benefit from
considering how techniques can be standardized to promote
maximum comparability of recordings made at different
sites and with different hardware.
3) In analogy to acoustic databases (where unlabelled data
is of only the most limited use), what segmentation and
labelling information (and tools for the exploitation
thereof) could/should be made available with the raw data
to facilitate flexible access for different purposes?
4) For future recordings, what categories of data and
corpora would be potentially of most widespread use?
If you are interested in participating please contact Phil
Hoole (preferably by Email) at the address below as soon as
possible. Further information on registration and format for
presentations will be distributed early in 1995.
Please also draw this letter to the attention of any
colleagues you think might be interested.
Phil Hoole and Hans Tillmann
Institut fuer Phonetik
Munich University
Schellingstr. 3
D - 80799 Munich
Germany
Fax: + 49 89 2800362
Email : HOOLE@SUN1.PHONETIK.UNI-MUENCHEN.DE