Sound direction estimation device, sound direction estimation method, and sound direction estimation program
A sound direction estimation device includes a transfer function storage unit configured to store transfer functions of sound sources in correlation with directions of the sound sources, a calculation unit configured to calculate the number of classes to be searched and a search interval for each class based on a desired search range and a desired spatial resolution for searching for the directions of the sound sources, and a sound source localization unit configured to search the search range for every search interval using the transfer function, to estimate the direction of the sound source based on the search result, to update the search range and the search interval based on the estimated direction of the sound source until the number of classes calculated by the calculation unit is reached, and to estimate the direction of the sound source.
1. A sound direction estimation device comprising:
a processor; and
microphones collecting sounds from sound sources,
wherein the processor is programmed with instructions that, when executed, cause the processor to function as:
a transfer function storage unit configured to store transfer functions of the sound sources in correlation with directions of the sound sources;
a calculation unit configured to calculate a number of classes to be searched and a search interval for each class based on a desired search range and a desired spatial resolution for searching for the directions of the sound sources; and
a sound source localization unit configured to search the search range for every search interval using a transfer function, to estimate the direction of at least one of the sound sources based on a search result, to update the search range and the search interval based on the estimated direction of the sound source until the number of classes calculated by the calculation unit is reached, and to estimate the direction of at least one of the sound sources,
wherein the sound source localization unit is configured
to determine whether the transfer function corresponding to an azimuth of the search interval is stored in the transfer function storage unit,
to read the transfer function corresponding to the azimuth of the search interval from the transfer function storage unit when it is determined that the transfer function corresponding to the azimuth of the search interval is stored in the transfer function storage unit,
to calculate an interpolated transfer function by interpolating the transfer function corresponding to the azimuth of the search interval by a Frequency Time Domain Linear or a Bi-linear Interpolation method when it is determined that the transfer function corresponding to the azimuth of the search interval is not stored in the transfer function storage unit, and
to estimate the directions of the sound sources using the read transfer function or the calculated interpolated transfer function,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method comprises calculating the interpolated transfer function with a Frequency Domain Linear or the Bi-linear Interpolation method and a Time Domain Linear Interpolation method,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method further comprises:
decomposing the calculated interpolation transfer function with the Frequency Domain Linear or the Bi-linear Interpolation method into a phase expression and an amplitude expression based on the Frequency Domain Linear or the Bi-linear Interpolation method; and
decomposing the calculated interpolation transfer function with the Time Domain Linear Interpolation method into another phase expression and another amplitude expression based on the Time Domain Linear Interpolation method,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method further comprises combining the phase expression decomposed based on the Frequency Domain Linear or Bi-linear Interpolation method with the another amplitude expression decomposed based on the Time Domain Linear Interpolation method, and
wherein the calculation unit is configured to calculate the number of classes (S) according to a following expression:
S
=
log
d
0
d
s
,
where d 0 represents the search range, d s represented the search interval of a given class s, and s is an integer representing the given class with a value between 1 and S, and wherein the search interval d s of the given class s is determined using the following expression:
d
s
=
d
0
S
-
s
d
S
s
s
=
d
0
S
-
s
S
d
S
s
S
,
so that a search number for each class is constant in all of the classes and a total search number in all the classes is minimized.
2. The sound direction estimation device according to claim 1 , wherein the sound source localization unit is configured
to search an n-th class (where n is an integer greater than or equal to 1) in a predetermined search range with the calculated search interval and update at least one search interval in the search range to a search range of an (n+1)-th class based on the search result,
to update a search interval of the (n+1)-th class to be searched based on the updated search range of the (n+1)-th class and the desired spatial resolution, and
to update and estimate the directions of the sound sources using the updated search range of the (n+1)-th class, the updated search interval of the (n+1)-th class, and the transfer function corresponding to an azimuth of the search interval until the number of classes (n+1) reaches the number of classes calculated by the calculation unit.
3. The sound direction estimation device according to claim 1 , wherein the calculation unit is configured to calculate the number of classes (S) according to the following expression:
S
=
log
(
d
0
d
s
∏
i
=
1
S
(
C
i
)
)
,
where C i represented a calculation cost of an a given i-th class, and i is an integer representing the given class with a value between 1 and S, and the search interval d s according to the following expression:
d
s
=
d
0
S
-
s
S
d
S
s
S
∏
i
=
1
S
(
C
i
S
-
s
s
)
∏
i
=
S
S
+
1
(
1
C
i
)
,
so as to minimize the calculation cost which is a sum of a search cost spent in calculating the search number in the search range and an interpolation cost spent in the interpolation.
4. A sound direction estimation method in a sound direction estimation device, comprising:
calculating a number of classes to be searched and a search interval for each class based on a desired search range and a desired spatial resolution for searching for a direction of a sound source;
searching the search range for every search interval using transfer functions from the sound source stored in a transfer function storage unit in correlation with the direction of the sound source and estimating the direction of the sound source based on a search result;
updating the search range and a search interval based on the estimated direction of the sound source until the calculated number of classes calculation unit is reached and estimating the direction of the sound source;
determining whether a transfer function corresponding to an azimuth of the search interval is stored in the transfer function storage unit,
reading the transfer function corresponding to the azimuth of the search interval from the transfer function storage unit when it is determined that the transfer function corresponding to the azimuth of the search interval is stored in the transfer function storage unit,
calculating an interpolated transfer function by interpolating the transfer function corresponding to the azimuth of the search interval by a Frequency Time Domain Linear or a Bi-linear Interpolation method when it is determined that the transfer function corresponding to the azimuth of the search interval is not stored in the transfer function storage unit, and
estimating the directions of the sound sources using the read transfer function or the calculated interpolated transfer function,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method comprises calculating the interpolated transfer function with a Frequency Domain Linear or the Bi-linear Interpolation method and a Time Domain Linear Interpolation method,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method further comprises:
decomposing the calculated interpolation transfer function with the Frequency Domain Linear or the Bi-linear Interpolation method into a phase expression and an amplitude expression based on the Frequency Domain Linear or the Bi-linear Interpolation method; and
decomposing the calculated interpolation transfer function with the Time Domain Linear Interpolation method into another phase expression and another amplitude expression based on the Time Domain Linear Interpolation method, and
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method further comprises combining the phase expression decomposed based on the Frequency Domain Linear or Bi-linear Interpolation method with the another amplitude expression decomposed based on the Time Domain Linear Interpolation method, and
wherein the calculation unit is configured to calculate the number of classes (S) according to a following expression:
S
=
log
d
0
d
s
,
where d 0 represents the search range, d s represented the search interval of a given class s, and s is an integer representing the given class with a value between 1 and S, and wherein the search interval d s of the given class s is determined using the following expression:
d
s
=
d
0
S
-
s
d
S
s
s
=
d
0
S
-
s
S
d
S
s
S
,
so that a search number for each class is constant in all of the classes and a total search number in all the classes is minimized.
5. A non-transitory computer readable medium storing a sound direction estimation program that causes a computer of a sound direction estimation device to perform:
calculating a number of classes to be searched and a search interval for each class based on a desired search range and a desired spatial resolution for searching for a direction of a sound source;
searching the search range for every search interval using transfer functions from the sound source stored in a transfer function storage unit in correlation with the direction of the sound source and estimating the direction of the sound source based on a search result;
updating the search range and a search interval based on the estimated direction of the sound source until the calculated number of classes calculation unit is reached and estimating the direction of the sound source;
determining whether a transfer function corresponding to an azimuth of the search interval is stored in the transfer function storage unit,
reading the transfer function corresponding to the azimuth of the search interval from the transfer function storage unit when it is determined that the transfer function corresponding to the azimuth of the search interval is stored in the transfer function storage unit,
calculating an interpolated transfer function by interpolating the transfer function corresponding to the azimuth of the search interval by a Frequency Time Domain Linear or Bi-linear Interpolation method when it is determined that the transfer function corresponding to the azimuth of the search interval is not stored in the transfer function storage unit, and
estimating the directions of the sound sources using the read transfer function or the calculated interpolated transfer function,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method comprises calculating the interpolated transfer function with a Frequency Domain Linear or the Bi-linear Interpolation method and a Time Domain Linear Interpolation method,
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method further comprises:
decomposing the calculated interpolation transfer function with the Frequency Domain Linear or the Bi-linear Interpolation method into a phase expression and an amplitude expression based on the Frequency Domain Linear or the Bi-linear Interpolation method; and
decomposing the calculated interpolation transfer function with the Time Domain Linear Interpolation method into another phase expression and another amplitude expression based on the Time Domain Linear Interpolation method, and
wherein the Frequency Time Domain Linear or the Bi-linear Interpolation method further comprises combining the phase expression decomposed based on the Frequency Domain Linear or Bi-linear Interpolation method with the another amplitude expression decomposed based on the Time Domain Linear Interpolation method, and
wherein the calculation unit is configured to calculate the number of classes (S) according to a following expression:
S
=
log
d
0
d
s
,
where d 0 represents the search range, d s represented the search interval of a given class s, and s is an integer representing the given class with a value between 1 and S, and wherein the search interval d s of the given class s is determined using the following expression:
d
s
=
d
0
S
-
s
d
S
s
s
=
d
0
S
-
s
S
d
S
s
S
,
so that a search number for each class is constant in all of the classes and a total search number in all the classes is minimized.