Digital Signal Processing Reference
In-Depth Information
Figure 3-3.
Experimental setup (Phase I).
3.2
Collection of Spoken Dialog Corpus - Phase I
3.2.1
The Experimental Setup
The spoken dialog corpus was recorded in a studio in Tokyo. The studio
consisted of two rooms as depicted in Figure 3-3. The left room represented a
car equipped with an in-vehicle terminal, whereas the right room represented
the remote server. In the left room, the speaker assumed the role of driver, and
in the right room, the operator assumed the role of the dialog system. The
speaker and the operator talked to each other using connected microphones
and headsets; the two rooms were completely separated so that no nonverbal
interactions took place between speaker and operator. The dialogs that took
place between the speaker and the operator were recorded on audiotape. The
operator answered queries from the speaker by acting as a travel agent with
access to the following information:
Real-time traffic information
Restaurant information
Sightseeing information
Hotel and
ryokan
(Japanese inn) information
In order to support task-oriented dialogs including detailed tourist
information, the operator role required expertise in travel information in
addition to the ability to speak correct Japanese. To meet these requirements,
we employed a professional female announcer with prior job experience in a
travel agency to act in the role of operator.
