Database Reference
In-Depth Information
the user when a recognition error occurs or an out-of-vocabulary is encountered. The
''watch-dog'' function monitors the service situation and status of the eASR/eTTS in
order to cope with the exception-handling error which can occur when a user pushes
the external buttons during the service interval.
Through the use of the speech interactive agent on embedded car navigation sys-
tem, next Section describes the interactive information generation method using sen-
sor and data fusion on the embedded system and then, the information gathering and
generation method via the Web.
3 Decision Theoretic Fusion Framework
This Section provides the base framework for embedded data mining from the raw
data to highly processed information. First, raw data is generalized by using the sen-
sor fusion method and then data fusion rules and active interaction rule generation is
employed. Finally, the embedded system is connected to the Web for effective data
processing and service and user satisfaction.
3.1 Sensory Fusion Rule
To perform the requests for speech interaction, firstly the sensory fusion model can be
expressed by
(
)
Y
=
f
O
/
K
,
Y
(1)
i
i
−
1
where
i
is a number of processing results,
O
is a observable sensory input,
K
is a domain
knowledge,
Y
i-1
is status information being processed from a previous time and
f()
is the
sensory fusion function to combine the sensory inputs and then control the current re-
quests given the previous situation. The observable sensor input,
O
is expressed by
k
(
)
(
)
( )
(
)
(
)
=
O
=
g
Mute
⋅
g
HF
⋅
g
R
⋅
g
Ptt
⋅
g
E
(2)
1
2
3
4
5
i
i
0
where
M
is a mute, HF is a hands-free,
R
is a remote controller,
Ptt
is a push-to-talk,
E
is a event created by service applications, and
k
is a number of applications being run
simultaneously. Each input is independent each other as well as processed parallelly.
The variable,
g()
is a function to observe and detect the sensor input. While a sensory
input between
g
1
and
g
4
is a direct input from a sensor,
g
5
is a transmitted input from
application programs via the inter-process communication. The sensory inputs can hap-
pen simultaneously. However, for the action to be performed promptly it is always one
function that is most suitable in a given situation. This is due to the fact that the hard-
ware resource has limitations, and the system can provide the robustness, consistency
and efficiency in using a service. Thus, the fusing function,
f()
should be considered
with respect to service quality and usability. In this paper, we apply the rule based deci-
sion function as a fusion function of respective inputs. In equation (1),
K
is a domain
specific knowledge to provide combing rules as shown in Table 1. The given rule is
decided by considering the service capability, priority and resource limitations, etc.
Decision categories are composed of five decision rules. By using this sensory fused
rule, data fusion rule is generalized for effective speech interaction in next subsection.
Search WWH ::
Custom Search