Methods and systems for training a machine learning model and authenticating a user with the model
A method of training a machine learning model to conduct authentication transactions is provided that includes the steps of obtaining, by an electronic device, a training dataset of audio signals. Each audio signal includes voice biometric data of a user and information for a passphrase spoken by the respective user and belongs to a same or different data class. Each data class includes a user identity and a passphrase identifier. Moreover, the method includes the steps of creating, using a machine learning model being trained, at least one embedding for each audio signal. The machine learning model includes parameters. Furthermore, the method includes calculating, by a machine learning algorithm using the embeddings, a loss, and updating parameters of the machine learning model based on the calculated loss. In response to determining criteria defining an end of training have been satisfied, deeming the machine learning model to be operable for use in simultaneously successfully verifying the identity of a user based on voice biometric data and verifying a passphrase spoken by the user matches a secret passphrase during authentication transactions.
1 . A method of training a machine learning model to conduct authentication transactions comprising the steps of:
obtaining, by an electronic device, a training dataset of audio signals, wherein each audio signal includes voice biometric data of a user and information for a secret passphrase spoken by the user and belongs to a same or different data class, each data class comprises a user identity and a passphrase identifier, and each secret passphrase is a unique passphrase chosen by a respective user and is known by the respective user only;
creating a feature matrix for each audio signal in each data class;
inputting the matrices into the machine learning model being trained, wherein the machine learning model being trained includes parameters;
calculating at least one embedding for each of the matrices;
calculating, by a machine learning algorithm using the embeddings, a loss for each data class and a second loss based on the losses for each data class;
updating the parameters of the machine learning model being trained based on the second loss; and
in response to determining criteria defining an end of training have been satisfied, deeming the machine learning model to be operable for use in simultaneously successfully verifying the identity of a user based on voice biometric data and verifying a passphrase spoken by the user matches the secret passphrase of the user during authentication transactions.
2 . The method according to claim 1 , further comprising, in response to determining criteria defining an end of training have not been satisfied, repeatedly conducting said creating, inputting, calculating at least one embedding, calculating a loss and a second loss, and updating steps until the criteria are satisfied.
3 . The method according to claim 1 , said calculating step comprising the step of calculating a metric loss, wherein the metric loss is a triplet loss, a prototypical loss, or a contrastive loss.
4 . The method according to claim 1 , said calculating step comprising calculating the loss according to the equations
L
s
=
1
n
s
(
n
s
-
1
)
∑
i
∈
N
s
∑
j
∈
N
s
\
i
max
(
f
(
x
s
i
)
-
f
(
x
s
j
)
2
2
-
f
(
x
s
i
)
-
f
(
x
r
k
)
2
2
+
α
trp
,
0
)
and
L
trp
=
1
n
∑
s
L
s
Where:
L trp is the triplet loss;
L s is the loss calculated separately for each class s;
Matrices x s i and x s j belong to the same class, class s, which means the user and the spoken passphrase for the matrices are the same;
Matrix x r k belongs to a different class, class r, at an occurrence k;
Functions ƒ(x s i ), ƒ(x s j ), ƒ( r k ) are embeddings extracted from the input matrices x s i , x s j and x r k , respectively;
N is an ensemble of different classes of data in the training dataset;
n is the number of elements in ensemble N;
N s is the ensemble of occurrences of audio signals for class s;
n s is the number of elements in ensemble s;
N s \i is the ensemble of occurrences of audio signals for class s except for occurrence i; and
α trp is a margin of error.
5 . An electronic device for training a machine learning model to conduct authentication transactions comprising:
a processor; and
a memory configured to store data, said electronic device being associated with a network and said memory being in communication with said processor and having instructions stored thereon which, when read and executed by said processor, cause said electronic device to:
obtain a training dataset of audio signals, wherein each audio signal includes voice biometric data of a user and information for a secret passphrase spoken by the user and belongs to a same or different data class, each data class comprises a user identity and a passphrase identifier;
create a feature matrix for each audio signal in each data class;
input the matrices into the machine learning model being trained, wherein the machine learning model being trained includes parameters;
calculate at least one embedding for each of the matrices;
calculate, by a machine learning algorithm using the embeddings, a loss for each data class and a second loss based on the losses for each data class;
update the parameters of the machine learning model being trained based on the second loss; and
in response to determining criteria defining an end of training have been satisfied, deem the machine learning model to be operable for use in simultaneously successfully verifying the identity of a user based on voice biometric data and verifying a passphrase spoken by the user matches the secret passphrase of the user during authentication transactions.
6 . The electronic device according to claim 5 , wherein the instructions when read and executed by said processor, cause said electronic device to repeatedly:
create a feature matrix for each audio signal in each class of data;
input the matrices into the machine learning model being trained;
calculate at least one embedding for each of the matrices;
calculate the loss for each class of data and the second loss; and
update the parameters of the machine learning model being trained based on the calculated losses until the criteria are satisfied.
7 . The electronic device according to claim 5 , wherein the instructions when read and executed by said processor, cause said electronic device to calculate a metric loss, wherein the metric loss is a triplet loss, a prototypical loss, or a contrastive loss.
8 . The electronic device according to claim 5 , wherein the instructions when read and executed by said processor, cause said electronic device to calculate the loss according to the equations
L
s
=
1
n
s
(
n
s
-
1
)
∑
i
∈
N
s
∑
j
∈
N
s
\
i
max
(
f
(
x
s
i
)
-
f
(
x
s
j
)
2
2
-
f
(
x
s
i
)
-
f
(
x
r
k
)
2
2
+
α
trp
,
0
)
and
L
trp
=
1
n
∑
s
L
s
Where:
L trp is the triplet loss;
L s is the loss calculated separately for each class s;
Matrices x s i and x s j belong to the same class, class s, which means the user and the spoken passphrase for the matrices are the same;
Matrix x r k belongs to a different class, class r, at an occurrence k;
Functions ƒ(x s i ), ƒ(x s j ), ƒ(x r k ) are embeddings extracted from the input matrices x s i , x s j and x r k , respectively;
N is an ensemble of different classes of data in the training dataset;
n is the number of elements in ensemble N;
N s is the ensemble of occurrences of audio signals for class s;
n s is the number of elements in ensemble s;
N s \i is the ensemble of occurrences of audio signals for class s except for occurrence i; and
α trp is a margin of error.
9 . A non-transitory computer-readable recording medium in an electronic device for training a machine learning model to conduct authentication transactions, the non-transitory computer-readable recording medium storing instructions which when executed by a hardware processor cause the hardware processor to perform steps comprising:
obtaining a training dataset of audio signals, wherein each audio signal includes voice biometric data of a user and information for a secret passphrase spoken by the user and belongs to a same or different data class, each data class comprises a user identity and a passphrase identifier;
creating a feature matrix for each audio signal in each data class;
inputting the matrices into the machine learning model being trained, wherein the machine learning model being trained includes parameters;
calculating at least one embedding for each of the matrices;
calculating, by a machine learning algorithm using the embeddings, a loss for each data class and a second loss based on the losses for each data class;
updating the parameters of the machine learning model being trained based on the second loss; and
in response to determining criteria defining an end of training have been satisfied, deeming the machine learning model to be operable for use in simultaneously successfully verifying the identity of a user based on voice biometric data and verifying a passphrase spoken by the user matches the secret passphrase of the user during authentication transactions.
10 . The non-transitory computer-readable recording medium according to claim 9 , wherein the instructions when read and executed by said hardware processor, cause said hardware processor to perform steps compromising calculating a metric loss, wherein the metric loss is a triplet loss, a contractive loss, or a prototypical loss.
11 . The non-transitory computer-readable recording medium according to claim 9 , wherein the instructions when read and executed by said processor, cause said hardware processor to repeatedly perform said creating, inputting, calculating at least one embedding, calculating the loss and the second loss, and updating steps until the criteria are satisfied.
12 . The non-transitory computer-readable recording medium according to claim 9 , wherein the instructions when read and executed by said processor, cause said hardware processor to perform steps compromising calculating the loss according to the equations
L
s
=
1
n
s
(
n
s
-
1
)
∑
i
∈
N
s
∑
j
∈
N
s
\
i
max
(
f
(
x
s
i
)
-
f
(
x
s
j
)
2
2
-
f
(
x
s
i
)
-
f
(
x
r
k
)
2
2
+
α
trp
,
0
)
and
L
trp
=
1
n
∑
s
L
s
Where:
L trp is the triplet loss;
L s is the loss calculated separately for each class s;
Matrices x s i and x s j belong to the same class, class s, which means the user and the spoken passphrase for the matrices are the same;
Matrix x r k belongs to a different class, class r, at an occurrence k;
Functions ƒ(x s i ), ƒ(x s j ), ƒ( r k ) are embeddings extracted from the input matrices x s i , x s j and x r k , respectively;
N is an ensemble of different classes of data in the training dataset;
n is the number of elements in ensemble N;
N s is the ensemble of occurrences of audio signals for class s;
n s is the number of elements in ensemble s;
N s \i is the ensemble of occurrences of audio signals for class s except for occurrence i and
α trp is a margin of error.