Digital Signal Processing Reference
In-Depth Information
can claim that the widest coverage of the user's utterance could be achieved
by using a LVCSR module and a statistical language model. In such a case,
an NLP module must be used to extract information about a specific attribute-
value pair. However, it is not reasonable to implement such modules in the
client system because the computational resources are limited in the vehicle.
Therefore, we use a simple speech recognizer with a regular grammar and a
small lexicon, defined as a <grammar>, in the ASR part of the client system.
Figure 5-5.
Transitions and grammars.
Another important role of a <grammar> is control of the dialog flow. Fig.
5-5 shows that each arc includes a <grammar>, specified by a filename with
the “.gram” extension. Since the VoiceXML specification allows us to include
multiple <grammar>s in a single form, we can easily split the flow of the
dialog by checking which <grammar> covers the user's utterance. If we have
the table of <grammar> names and the table of attribute-value pairs allowed
