 |
DaimlerChrysler Research and Technology |
SPEECH UNDERSTANDING SYSTEMS
For human beings language is the most important means of communication.
But far beyond this it also forms the basis of our thought processes. We
use language to formulate concepts, to express relationships between them,
and to build up our store of human knowledge.
The future of information science will therefore be characterized by intensive
use of natural language and speech for man machine interaction and by advanced
automatic assistant systems for handling human knowledge, which is mainly
based on language. Machines will automatically recognize the words of human
speech, understand what is being said, and give answers in synthetically
generated speech. Thus they will have access to verbal information and
be able to assist us in a dialog with their specific capabilities.
The speech understanding group of Daimler-Benz is since many years active
in research in all these fields of speech understanding and speech and
language processing. The main areas of these activities are short described
in the following documents.
Automatic Speech Recognition and Linguistic
Analysis
Automatic recognition of spoken words is the prerequisite for a large number
of applications. Simple systems capable of recognizing a few hundred words
can be used for such things as operating entertainment in the car, appliances,
simple navigation systems in the car and interactive voice response systems
over the telephone. With systems that can recognize several thousand words
it becomes possible to realize things like a voice-actuated typewriter
where the user simply dictates what he wants written down.
Word recognition is however only the first of several steps whose ultimate
goal is to identify the meaning of what is being said, thus making true
information processing of speech possible. Intensive research in linguistic
analysis of spoken sentences or at least phrases is necessary to analyze
the syntax and the semantics. Finally of course this has to be integrated
into the pragmatics and applications of speech dialogue systems. Especially
the linguistic analysis of spontaneously spoken sentences has to care about
the analysis of even syntactically incorrect parts and it should in spite
of this correctly extract the meaning. Automatic Speech Generation and
Synthesis
The results of speech recognition must be verbally confirmed. Automatic
speech generation and synthesis of spoken language makes this possible.
The necessary process has first to generate a linguistic phrase from the
semantic description and afterwards this phrase must be transferred into
a speech signal. Reading machines can be constructed which make it possible
to get spoken information from documents transmitted over e-mail or the
Internet or from a databank. One can even build reading machines for the
blind, or -what will become the most important thing - generate speech
for response from a dialog system.Dialogue Systems
In our research activities we are also studying the problem of how to construct
man-machine dialogues. Human beings converse in a very free form whereas
machines prefer very rigidly structured dialogues. Here an acceptable compromise
between these two extremes must be found for the man-machine dialogue.
The long range goal is of course to make the free form of human beings
accessible to the machine.
InfoPort
At the bottom line applications utilizing speech-understanding systems
play a critical role in our research activities. The aim of the InfoPort
Project is to show what future possibilities will be opened up when an
integrated approach is taken to processing natural language information
in both spoken and written form using the latest tools in information science.
The goal here is to correctly classify the profusion of information which
we receive daily and pass on more or less modified and processed, make
it easily accessible, and also to supplement our limited human memory.
InfoPort may be thought of as a personal assistance system which evaluates
all incoming channels of natural language information, generates cross
references, stores relevant information, and then finally makes this information
available in spoken and written form. Dialogue elements ease the access
to this information. InfoPort, the integrated information assistant, points
the way to the future of information science.
List of all the group members
, including their phone and room numbers, E-mail addresses and personal
Home Pages.
Recent publications
of the group.
Speech Understanding Group
Daimler Benz Research & Technology
Wilhelm-Runge-Str. 11
89081 - Ulm - Germany
Phone: +49 - 731 - 505 -2142
Fax: +49 - 731 - 505- 41 05