Digital Signal Processing Reference
In-Depth Information
From Fig. 5-7, we see that steps (1) to (5) represent the data flow sequence
of a typical route guidance dialog. This architecture has a distinct advantage
when compared to standard client-server dialog systems, since it can continue
the route guidance dialog even if the communication between the client and
the server is disconnected. In addition, the DM can provide dynamic
information to the user whenever the connection to the remote server is
established.
The new dialog (i. e. parking dialog) must be aborted by a trigger, because
the only mechanism for embedding dialogs in VoiceXML is via the
<subdialog> directive, which pushes the original dialog (e. g. directions
dialog) down on a dialog stack and invokes a new dialog. As a result, when
the user re-activates a previous (parking) dialog after the trigger was
processed in the original dialog, the default VoiceXML action would be to
push a new dialog and start it in its initial state (rather than returning to
whatever the last active state was in a previously-activated dialog). However,
if the server uses session variables to maintain a history of suspended dialogs
and their last active state, it is possible to restart a dialog from the point at
which it had been suspended. This capability represents an enhancement to
our prototype system, and is under active development.
7.
CONCLUSIONS
In this chapter, we described a dialog management architecture for car
telematics systems. The architecture consists of a client and a server, and is
designed to minimize the bandwidth of communications between them. On
the server side, the Dialog Manager controls the current interaction with the
user according to pre-defined scenarios written in ScenarioXML. The
developer can define state-and-transition dialog scenarios using various
predefined templates, which support integration of dynamic information from
external databases and information services. Analysis of each user utterance
within a dialog is achieved through application of a pre-selected grammar and
lexicon, so that the developer has only to select appropriate sets of grammars
and lexicons for each dialog state. These grammars may be written by hand,
but it is also possible to construct them automatically using sample dialogs for
each task. Finally, we described how the system switches control between the
server and the client according to the current status of the communication
channel. The system is robust in the presence of sudden network
disconnections, and bandwidth utilization can be reduced by the use of client-
side applications for simple tasks which do not require real-time access to
dynamically changing information.
