Digital Signal Processing Reference
In-Depth Information
dialogue and keeping it going smoothly. It plays an important role in an
MSDS that is described in the next section.
Figure 4-1.
Basic components of a multi-speaker dialogue system. The rectangle blocks are the
processing units; the parallelograms are multiple outputs derived from the processing units.
3.
DIALOGUE MANAGEMENT FOR MSDS
Once the active speaker is determined, the target speech is sent to the
speech recognition and natural language processing components. The
keyword spotting and partial parsing techniques that are popular in the field
of spoken language processing can be adopted in an MSDS. The parsed result
will be the most likely words sequence with their part-of-speech tags. They
are then fed to the dialogue manager. The dialogue manager maintains the
interaction among the system and the multiple speakers and keeps it going
smoothly. Fig.
4-1
shows the block diagram of the multi-speaker dialogue
management.
In an MSDS, each speaker may have his own individual goal for
information retrieval. In contrast to the individual goal, the global goal is the
integration of each individual goal. The management of the multi-speaker
dialogue has several functions: 1) to interpret intentions and semantics of each
individual speaker in order to detect if there is a conflict between speakers, 2)
to integrate individual goals into global goals, 3) to determine whether a
specific goal is completed, and 4) to generate the response. In this section, we
illustrate how the management of MSDS works by giving an algorithm and
some examples.
