Database Reference
In-Depth Information
Table 1.
The negotiation rule table according to the priority control
Current
State
Previous
State
eASR is
requested
Application TTS
is requested
CNS TTS is
requested
Hands-
Free But-
ton pushed
Mute
Button
pushed
Hands-Free
button
enable
Disabled
Disabled
Enabled
Not appli-
cable
Not appli-
cable
Mute button
enable
Enabled
Disabled
Enabled
Not appli-
cable
Not appli-
cable
eASR run-
ning
Previous
eASR exits
and new
eASR runs
Previous eASR
exits and eTTS
starts
eASR runs
continuously
and CNS TTS
starts
eASR
exits
eASR
exits
Application
eTTS run-
ning
Previous
eTTS stops
and
Previous eTTS
stops and new
eTTS starts
Application
eTTS pauses
and CNS TTS
starts
eTTS
stops
eTTS
stops
eASR
runs
CNS eTTS
running
CNS eTTS
starts and
eASR runs
Previous CNS
eTTS finishs
and then appli-
cation eTTS
starts
Previous CNS
eTTS stops
and new CNS
eTTS starts
Don't care
Don't care
3.2 Data Fusion Rules from Interface Modalities
When given the sensory fusion result, the speech agent can decide the action to be
performed. Next, the data fusion model for speech interaction can be expressed by
()()()
(3)
Z
=
H
O
⋅
I
P
⋅
J
Y
,
i
=
1
...,
3
i
i
() (
)
(4)
H
O
=
h
O
/
M
,
i
=
1
,...,
3
i
i
i
i
i
where
i
is the number of speech interaction tools and
H
i
(O
i
)
is a speech interaction
tool; 1)embedded speech recognition, 2)distributed speech recognition 3)text-to-
speech. Thus, the variable,
O
1
and
O
2
are speech sampling data and
O
3
is text data.
Thus,
H
i
(O
i
)
is decomposed as follows.
H
O
h
O
/
M
1
1
1
1
1
(5)
arg
max
#
W
L
O
/
W
k
j
j
where
h
1
(O
1
)
is a pattern recognizer using the maximum a posteriori (MAP) decision
rule to find the most likely sequence of words.
()
(6)
H
O
=
h
(
O
/
M
)
=
h
(
O
)
2
2
2
2
2
2
2
where
h
2
(O
2
)
is a front-end feature extractor to pass the speech features into the back-
end distributed speech recognition server.
Search WWH ::
Custom Search