From: dermotfcampbell <dermotfcampbell@...>
Subject: [iatefl_pronsig] Update on DIT's Dynamic Speech Corpus (under development)
To: iatefl_pronsig@...
Received: Wednesday, 19 November, 2008, 1:28 AM

Hi Sue and Chris,

I am providing an outline of what we propose to talk about at IATEFL
2009 (if accepted!). It concerns the nascent Dublin Institute of
Technology Dynamic Speech Corpus the development of which is funded
by Enterprise Ireland under a 30 month project called FLUENT, and
which is (not surprisingly given the source of the funding) destined
to be a commercial product – that is why I can't go into technical
detail.
The Dynamic Speech Corpus (DSC) is aimed at autonomous learners,
teachers, authors and researchers – in that order. It is a unique
audio resource based on high quality recordings (4 times CD quality)
of natural dialogues. The reason for the high-end recording
technology is to allow instrumental analysis of the audio assets as
time and resources permit.
The idea started from the wish to have a speaking concordancer: line
up the hits from a character string search and actually listen to
the results. Most concordancers we came across were fixated on
written texts or at most the transcripts of speech – but almost
never allowed access to the original recorded speech and never to
natural dialogues. Others were fixated on `authenticity' and the
audio quality suffered as well. It's difficult to analyse authentic
noise!
Our plan is to produce a tool which allows learners to realize
Richard Cauldwell's dream: that learners should `spend more time
with the signal'. The corpus will therefore be populated with
authentic dialogues (we call them `free-rangers' ), characterized by
all the `imperfections' of real speech, including the native speaker
(NS) stratagems of elision, assimilation, vowel centralization etc. –
often influenced by speed or speaker convenience.
We want to develop a small corpus of as-natural-as- it-gets dialogues
which will allow learners etc to find samples of NS-NS speech which
can prove difficult for learners to cope with, study the phonetic
and pragmatic environment in which the recording was made (a dynamic
interaction between two speakers) – and then slow down the sample,
if desired, so as to study the prosody of the speech and make the
segmental and suprasegmental aspects of the speech accessible to the
user. It's easier to attend to what you actually hear – and produce,
if that is desired.
Most of our effort at the moment goes into the architecture of the
necessary software, but in parallel we are prioritizing linguistic
tags for inclusion in the database. Prototype completion date: July
2010. I'll post it in this forum, if we get accepted by IATEFL.
WANG Yi and I have a shared interest in studying the formulaic
sequences we come across in our recordings – particularly from a
prosodic perspective, and we look forward to publishing on that
topic when we have enough data. But first she has to finish her PhD
and I have a database to help populate!

Dermot