Signal processing apparatus and method
The present technology relates to a signal processing apparatus and a method that enable improvement in the robustness of emotion estimation against noise. A signal processing apparatus extracts, on the basis of a measured biological signal, a physiological measure contributing to an emotion as a feature amount, outputs, with respect to time-series data about the feature amount, time-series data about a prediction label of an emotion status by a discriminative model built in advance, and outputs an emotion estimation result on the basis of a result of performing weighted summation of the prediction label with prediction label reliability that is reliability of the prediction label. The present technology can be applied to an emotion estimation processing system.
1 . A signal processing apparatus, comprising:
a biological sensor that measures a biological signal of a living body;
a second sensor that measures a body movement signal representative of changes in body movement of the living body;
a feature amount extraction unit operably coupled to the biological sensor and the second sensor, and that extracts, on a basis of the biological signal measured by the biological sensor, a physiological measure contributing to an emotion as a feature amount, wherein to reduce body movement noise in the biological signal measured by the biological sensor, the feature amount extraction unit extracts the physiological measure further on a basis of the body movement signal measured by the second sensor;
an emotion status time-series labelling unit that outputs, with respect to time-series data about the feature amount, time-series data about a prediction label of an emotion status by a discriminative model built in advance;
a signal quality determination unit that determines a signal quality of the biological signal; and
a stabilization processing unit that outputs an emotion estimation result on a basis of a result of performing a weighted summation of the prediction label with a prediction label reliability and a determination result of the signal quality, wherein the prediction label reliability is a reliability of the prediction label, wherein the stabilization processing unit calculates a measure-of-central-tendency reliability that is a reliability of a representative value of the prediction label by performing a weighted summation of the prediction label with the prediction label reliability and the determination result of the signal quality and outputs the representative value of the prediction label as the emotion estimation result by threshold processing on the measure-of-central-tendency reliability, and wherein the stabilization processing unit calculates the measure-of-central-tendency reliability in accordance with the expression
r
(
t
)
=
∑
i
w
i
c
i
s
i
y
i
Δ
t
i
/
∑
i
w
i
Δ
t
i
where y denotes the prediction label, c denotes the prediction label reliability, s denotes a signal quality score that is the determination result of the signal quality, and Δt i denotes a continuation time of an i-th event.
2 . The signal processing apparatus according to claim 1 , wherein the emotion status time-series labelling unit outputs, with respect to the time-series data about the feature amount in a sliding window, the time-series data about the prediction label of the emotion status in the sliding window by the discriminative model.
3 . The signal processing apparatus according to claim 1 , wherein the signal quality determination unit outputs the signal quality score for each class m in accordance with the following expression by using two or more types of class labels discriminated by a discriminative model for quality determination, reliability d of each of the class labels, and a function f( ) for adjusting the reliability d
s
m
=
α
m
f
(
d
m
)
.
4 . The signal processing apparatus according to claim 3 , wherein as to one or more types of respective classes, the function f( ) monotonically increases with respect to a class having the signal quality discriminated to be better than a predetermined threshold and monotonically decreases with respect to a class having the signal quality discriminated to be worse than the predetermined threshold and contain noise.
5 . The signal processing apparatus according to claim 1 , wherein the signal quality determination unit outputs the signal quality score depending on a high/low of signal periodicity.
6 . The signal processing apparatus according to claim 1 , wherein the stabilization processing unit calculates, in a case where the feature amount extracted from j-types of biological signals is set to be an input variable, the measure-of-central-tendency reliability of the prediction label in a sliding window in accordance with the expression and on a basis of a result of performing a weighted summation of the signal quality score for each of the j-types of the biological signals.
7 . The signal processing apparatus according to claim 6 , wherein assuming that w jk denotes a degree of model contribution of a feature amount k belonging to the j-types of biological signals, a weight Wj in a weighted-sum calculation of the signal quality score is expressed in accordance with the following expression
W
j
=
∑
k
∈
F
j
❘
"\[LeftBracketingBar]"
w
jk
❘
"\[RightBracketingBar]"
,
and
is calculated in accordance with the following expression
s
m
=
∑
j
W
j
α
m
,
j
f
(
d
m
,
j
)
/
∑
j
W
j
.
8 . The signal processing apparatus according to claim 1 , wherein the biological signal is a signal obtained by measuring at least one of an electroencephalogram, emotional sweating, a pulse wave, blood flow, or a continuous blood pressure.
9 . The signal processing apparatus according to claim 1 , further comprising a third sensor operably coupled to the feature amount extraction unit and that measures a pressure signal representative of changes in pressure of the living body in a detection area of the biological sensor, wherein to reduce body movement noise in the biological signal measured by the biological sensor, the feature amount extraction unit extracts the physiological measure further on a basis of the pressure signal measured by the third sensor.
10 . The signal processing apparatus according to claim 1 , wherein a casing is configured to be wearable.
11 . A signal processing method, comprising:
by a signal processing apparatus,
extracting, on a basis of a biological signal of a living body measured by a biological sensor, a physiological measure contributing to an emotion as a feature amount, wherein the extracting comprises reducing body movement noise in the biological signal by extracting the physiological measure further on a basis of a body movement signal of the living body measured by a second sensor;
outputting, with respect to time-series data about the feature amount, time-series data about a prediction label of an emotion status by a discriminative model built in advance;
determining a signal quality of the biological signal; and
outputting an emotion estimation result on a basis of a result of performing a weighted summation of the prediction label with a prediction label reliability and a determination result of the signal quality, wherein the prediction label reliability is a reliability of the prediction label, and wherein outputting the emotion estimation result comprises:
calculating a measure-of-central-tendency reliability that is a reliability of a representative value of the prediction label by performing a weighted summation of the prediction label with the prediction label reliability and the determination result of the signal quality and outputting the representative value of the prediction label as the emotion estimation result by threshold processing on the measure-of-central-tendency reliability, wherein the calculating comprises calculating the measure-of-central-tendency reliability in accordance with the expression
r
(
t
)
=
∑
i
w
i
c
i
s
i
y
i
Δ
t
i
/
∑
i
w
i
Δ
t
i
where y denotes the prediction label, c denotes the prediction label reliability, s denotes a signal quality score that is the determination result of the signal quality, and Δt i denotes a continuation time of an i-th event.
12 . The signal processing method according to claim 11 , wherein the reducing comprises reducing the body movement noise in the biological signal by extracting the physiological measure further on a basis of a pressure signal representative of changes in pressure of the living body in a detection area of the biological sensor measured by a third sensor.
13 . The signal processing method according to claim 11 , wherein, in a case where the feature amount extracted from j-types of biological signals is set to be an input variable, the calculating comprises calculating the measure-of-central-tendency reliability of the prediction label in a sliding window in accordance with the expression
and on a basis of a result of performing weighted summation of the signal quality score for each of the j-types of the biological signals, and wherein the method further comprises calculating a weight Wj in a weighted-sum calculation of the signal quality score in accordance with the following expression
s
m
=
∑
j
W
j
α
m
,
j
f
(
d
m
,
j
)
/
∑
j
W
j
,
wherein Wj is expressed in accordance with the following expression
W
j
=
∑
k
∈
F
j
❘
"\[LeftBracketingBar]"
w
jk
❘
"\[RightBracketingBar]"
,
and wherein w jk denotes a degree of model contribution of a feature amount k belonging to the j-types of biological signals.
14 . The signal processing method according to claim 11 , further comprising outputting, with respect to the time-series data about the feature amount in a sliding window, the time-series data about the prediction label of the emotion status in the sliding window by the discriminative model.
15 . The signal processing method according to claim 11 , further comprising outputting the signal quality score depending on a high/low of signal periodicity.