|
PlainBabel – Simultaneous Machine Translation
Until now machine speech-to-speech
translation was merely a vision for the future that wasn’t realizable
with adequate quality. Pooling the years of experience from researchers at
the Apache University of Technology and AppTek, Inc in Mclean, VA has
managed to combine intelligent individual modules for speech recognition,
translation and synthesis to create a system that realizes this vision.
With PlainBabel they have created a dialogue driven speech-to-speech
translation
system that allows for a bilingual natural dialogue between two parties that
don’t speak
each others language. The system can be used with different types of output
devices
from an ISDN telephone to a 3G mobile phone.
The experts from AppTek use intelligent systems based on statistical methods
for
speech recognition and translation. Unlike classical speech recognition and
translation
systems they do not rely on narrowly defined rules and vocabularies, but
rather the
system is trained with bilingual materials from the domain for which it will
be deployed.
Depending on the scope and specification of the training materials, the
system can
deliver a level of quality that was unattainable with previous systems. The
statistical
approach minimizes the increase of errors combining speech recognition and
machine
translation, allowing a speaker independent dialogue between two parties
within a
defined domain. Communication is supported and guided by a dialogue
management
system developed specifically for this purpose.
Figure 1: PlainBabel Server
|
|
