Digital Signal Processing Reference
In-Depth Information
2.
SYSTEM ARCHITECTURE
Figure 3-1 depicts the current system architecture of CAMMIA, which
consists of in-vehicle terminals and the server which are connected by a
wireless network. The client consists of an ASR, a VoiceXML interpreter,
and a TTS, whereas the server consists of a DM, a database, and a set of
dialog scenarios. The DM makes a VoiceXML text according to a dialog task
using a Dialog Scenario database and delivers it to an in-vehicle terminal. In
the in-vehicle terminal, the VoiceXML interpreter receives the VoiceXML
text and real dialog interactions between a used and the system can be carried
over. A spoken dialog corpus was used to evaluate the lexicon and grammar
of ASR and suitable dialog scenarios that represent particular dialog tasks
such as traffic information. The corpus was also used to evaluate the system's
coverage of different dialog tasks; these evaluations are discussed in Section 5.
Figure 3-1.
System Architecture of CAMMIA.
3.
COLLECTION OF SPOKEN DIALOG CORPUS
The collection of a spoken dialog corpus consists of three steps: speaker
selection, experimental setup, and task configuration. The most difficult and
important thing is to encourage the speakers to interact spontaneously and
effectively. The experimental setup and task configuration have a significant
impact on their ability to do so. This section focuses on the experimental
setup and task configuration which was employed.
