the MMDAgent as a voice navigation model. The proposed
system separates the entire navigation route into the
sections between one intersection and the next intersection
to turn right or left. We call each section a STEP section.
The proposed system navigates in the following order:
1) Departure point.
2) Repeat to navigate each STEP section.
3) Destination.
In each STEP section, the proposed system navigates in the
following order:
1) Path navigation
2) Intersection navigation
3) Instruction to change direction
Here, let us regard each navigation as a transition state, and
the end of each navigation or an utterance of users as a
transition condition. That is, we consider that navigation
is similar to state transition. In the proposed model, each
STEP section comprises multiple transition states, such as
path navigation and intersection navigation. Furthermore,
the intersection navigation also comprises state transitions.
Consequently, the state transition is very complicated. We
therefore define states of each guide in this model as FST
templates, navigation transitions are simplified because each
state has specific significance, such as the state of the STEP
section navigation and stated of the intersection navigation.
The voice navigation model is depicted in Figs. 6 and 7.