Digital Signal Processing Reference
In-Depth Information
defines a task, and the dialog manager can activate one or more tasks by
enumerating a specific set of (grammar, lexicon) pairs. Using this approach, a
developer can develop a robust dialog system for a particular task in a
straightforward manner. Such grammars and lexicons may be created by
hand, or derived from corpora that exemplify typical sentences from the
specified task.
The third challenge addressed by our system is task switching between the
client and the server systems. Sometimes the in-vehicle client loses its
connection to the remote server, and must continue to interact with the user as
a stand-alone system. In such situations, the client system continues the
dialog in a reduced way, providing limited information to the user. After the
connection is re-established, the client and the server negotiate to synchronize
the current dialog context, and start or continue the next task as appropriate.
Figure 5-1.
System architecture.
SYSTEM ARCHITECTURE
2.
Figure 5-1 illustrates the architecture of the car telematics system
described in this chapter. In the client system, the VoiceXML Interpreter
interacts with the user via automatic speech recognition (ASR) and text-to-
speech (TTS) interfaces. We use the VoiceXML Interpreter developed by
Hitachi CRL[8]; it supports most of the functions defined in VoiceXML
2.0[1], and also includes an additional input/output channel to allow
asynchronous communication with an internal application such as a GPS
navigator. Therefore, although the system usually communicates by sending
