Speechdat-AT databases of Austrian German telephone speech
What is it?
Speechdat-AT is a pair of telephone speech databases of Austrian German,
one from the wireline and one from the wireless (GSM EFR) network. The
databases were compiled by Speech&More for our partner companies,
and are now also distributed to the public, subject to agreement by our
project partners in each individual case. Speakers in the databases
were chosen to assure a representative distribution over accent regions,
sex, and age groups. The databases are compliant with the guidelines of
the Speechdat project, which have
become a European de-facto standard for speech databases. A successful
validation of Speechdat-AT was performed by
SPEX, Nijmegen.
Who needs it?
Anyone who wants to deploy a speaker-independent speech recogniser for use
via the telephone network in Austria. Speechdat-AT is currently the only
major speech database of Austrian German that exists. Without it, you would have
to use a speech recogniser trained on speech data from Germany. Philips
Speech Processing, who have incidentally used Speechdat-AT for building an
Austrian version of their FreeSpeech dictation software, have found in
an evaluation study that a recogniser trained with Speechdat-AT had a 30%
lower word error rate on Austrian German data than a recogniser trained
on data from Germany.
How much does it cost?
We provide licenses for Speechdat-AT at prices comparable to those of
the
European Linguistic
Resources Association (ELRA) for similar databases
for other languages. Contact us for a quote. Alternatively, joining the
Speech&More project will give you the chance to obtain Speechdat-AT for a
much reduced fee.
Reference
|