Method and apparatus for speaker identification using cepstral covariance matrices and distance metrics
View Patent ↗Disclosed is a method of automated speaker identification, comprising receiving a sample speech input signal from a sample handset; deriving a cepstral covariance sample matrix from the first sample speech signal; calculating, with a distance metric, all distances between the sample matrix and one or more cepstral covariance signature matrices; determining if the smallest of the distances is below a predetermined threshold value; and wherein the distance metric is selected from d 5 ( S , Σ ) = A + 1 H - 2 , d 6 ( S , Σ ) = ( A + 1 H ) ( G + 1 G ) - 4 , d 7 ( S , Σ ) = A 2 H ( G + 1 G ) - 1 , d 8 ( S , Σ ) = ( A + 1 H ) ( G + 1 G ) - 1 , d 9 ( S , Σ ) = A G + G H - 2 , fusion derivatives thereof, and fusion derivatives thereof with d 1 ( S , Σ ) = A H - 1.
1. A method of automated speaker identification, comprising:
receiving a sample speech input signal from a sample handset;
deriving a cepstral covariance sample matrix from said first sample speech signal;
calculating, with a distance metric, all distances between said sample matrix and one or more cepstral covariance signature matrices;
determining if the smallest of said distances is below a predetermined threshold value; and
wherein said distance metric is selected from
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
fusion derivatives thereof, and
fusion derivatives thereof with
d
1
(
S
,
Σ
)
=
A
H
-
1.
2. The method of claim 1 further comprising:
identifying said sample handset;
identifying a training handset used to derive each said signature matrix;
wherein for each said signature matrix, an adjusted sample matrix is derived by adding to said sample matrix a distortion matrix comprising distortion information for said training handset used to derive said signature matrix; and
wherein for each signature matrix, an adjusted signature matrix is derived by adding to each said signature matrix a distortion matrix comprising distortion information for said sample handset.
3. The method of claim 2 , wherein the step of identifying said sample handset further comprises:
calculating, with a distance metric, all distances between said sample matrix and one or more cepstral covariance handset matrices, wherein each said handset matrix is derived from a plurality of speech signals taken from different speakers through the same handset; and
determining if the smallest of said distances is below a predetermined threshold value.
4. The method of claim 3 wherein said distance metric satisfies symmetry and positivity conditions.
5. The method of claim 4 , wherein said distance metric is selected from
d
1
(
S
,
Σ
)
=
A
H
-
1
,
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
and
fusion derivatives thereof.
6. The method of claim 2 , wherein the step of identifying said training handset for each said signature matrix further comprises:
calculating, with a distance metric, all distances between said signature matrix and one or more cepstral covariance handset matrices, wherein each said handset matrix is derived from a plurality of speech signals taken from different speakers through the same handset; and
determining if the smallest of said distances is below a predetermined threshold value.
7. The method of claim 6 wherein said distance metric satisfies symmetry and positivity conditions.
8. The method of claim 7 , wherein said distance metric is selected from
d
1
(
S
,
Σ
)
=
A
H
-
1
,
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
and
fusion derivatives thereof.
9. A method of automated speaker identification, comprising:
receiving a sample speech input signal from a sample handset;
deriving a cepstral covariance sample matrix from said first sample speech signal;
calculating, with a distance metric, all distances between an adjusted sample matrix and one or more adjusted cepstral covariance signature matrices, each said signature matrix derived from training speech signals input from a training handset;
determining if the smallest of said distances is below a predetermined threshold value;
wherein for each said signature matrix, said adjusted sample matrix is derived by adding to said sample matrix a distortion matrix comprising distortion information for said training handset used to derive said signature matrix; and
wherein each said adjusted signature matrix is derived by adding to each said signature matrix a distortion matrix comprising distortion information for said sample handset.
10. The method of claim 9 , wherein said distance metric satisfies symmetry and positivity conditions.
11. The method of claim 10 , wherein said distance metric is selected from
d
1
(
S
,
Σ
)
=
A
H
-
1
,
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
and
fusion derivatives thereof.
12. The method of claim 9 , wherein said sample handset is identified by a method comprising:
calculating, with a distance metric, all distances between said sample matrix and one or more cepstral covariance handset matrices, wherein each said handset matrix is derived from a plurality of speech signals taken from different speakers through the same handset; and
determining if the smallest of said distances is below a predetermined threshold value.
13. The method of claim 12 , wherein said distance metric satisfies symmetry and positivity conditions.
14. The method of claim 13 , wherein said distance metric is selected from
d
1
(
S
,
Σ
)
=
A
H
-
1
,
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
and
fusion derivatives thereof.
15. The method of claim 9 , wherein, for each said signature matrix, said training handset is identified by a method comprising:
calculating, with a distance metric, all distances between said signature matrix and one or more cepstral covariance handset matrices, wherein each said handset matrix is derived from a plurality of speech signals taken from different speakers through the same handset; and
determining if the smallest of said distances is below a predetermined threshold value.
16. The method of claim 15 , wherein said distance metric satisfies symmetry and positivity conditions.
17. The method of claim 16 , wherein said distance metric is selected from
d
1
(
S
,
Σ
)
=
A
H
-
1
,
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
and
fusion derivatives thereof.
18. A program storage device, readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for automated speaker identification, said method steps comprising:
receiving a sample speech input signal from a sample handset;
deriving a cepstral covariance sample matrix from said first sample speech signal;
calculating, with a distance metric, all distances between said sample matrix and one or more cepstral covariance signature matrices;
determining if the smallest of said distances is below a predetermined threshold value; and
wherein said distance metric is selected from
d
5
(
S
,
Σ
)
=
A
+
1
H
-
2
,
d
6
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
4
,
d
7
(
S
,
Σ
)
=
A
2
H
(
G
+
1
G
)
-
1
,
d
8
(
S
,
Σ
)
=
(
A
+
1
H
)
(
G
+
1
G
)
-
1
,
d
9
(
S
,
Σ
)
=
A
G
+
G
H
-
2
,
fusion derivatives thereof, and
fusion derivatives thereof with
d
1
(
S
,
Σ
)
=
A
H
-
1.
19. A program storage device, readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for automated speaker identification, said method steps comprising:
receiving a sample speech input signal from a sample handset;
deriving a cepstral covariance sample matrix from said first sample speech signal;
calculating, with a distance metric, all distances between an adjusted sample matrix and one or more adjusted cepstral covariance signature matrices, each said signature matrix derived from training speech signals input from a training handset;
determining if the smallest of said distances is below a predetermined threshold value;
wherein for each said signature matrix, said adjusted sample matrix is derived by adding to said sample matrix a distortion matrix comprising distortion information for said training handset used to derive said signature matrix; and
wherein each said adjusted signature matrix is derived by adding to each said signature matrix a distortion matrix comprising distortion information for said sample handset.