Limiting identity space for voice biometric authentication
Disclosed are systems and methods including computing-processes executing machine-learning architectures extract vectors representing disparate types of data and output predicted identities of users accessing computing services, without express identity assertions, and across multiple computing services, analyzing data from multiple modalities, for various user devices, and agnostic to architectures hosting the disparate computing service. The system invokes the identification operations of the machine-learning architecture, which extracts biometric embeddings from biometric data and context embeddings representing all or most of the types of metadata features analyzed by the system. The context embeddings help identify a subset of potentially matching identities of possible users, which limits the number of biometric-prints the system compares against an inbound biometric embedding for authentication. The types of extracted features originate from multiple modalities, including metadata from data communications, audio signals, and images. In this way, the embodiments apply a multi-modality machine-learning architecture.
1. A computer-implemented method comprising:
generating, by a computer, one or more enrolled biometric-prints for an enrollee-user in an enrollee-record of a database, by applying one or more biometric recognition engines of a machine-learning architecture on enrollment biometric data for the enrollee-user;
receiving, by the computer, a first authentication request from a first provider server to authenticate a first inbound user, and a second authentication request from a second provider server to authenticate a second inbound user;
generating, by the computer, a first inbound biometric-print and a second inbound biometric-print corresponding to the first inbound user and the second inbound user, by applying the one or more biometric engines on corresponding first inbound contact data and second inbound contact data; and
transmitting, by the computer, a first result for a first computing service to the first provider server based upon a similarity between the first inbound biometric-print and an enrolled biometric-print of the one or more enrolled biometric-prints in the enrollee-record in accordance with a first biometric-print similarity threshold determined based on the first inbound contact data, and a second result for a second computing service to the second provider server based upon the similarity between the second inbound biometric-print and the enrolled biometric-print of the one or more enrolled biometric-prints in the enrollee-record in accordance with a second biometric-print similarity threshold determined based on the second inbound contact data.
2. The method according to claim 1 , further comprising updating, by the computer, the enrollee-record to indicate one or more computing services associated with the enrollee-user.
3. The method according to claim 1 , further comprising:
obtaining, by the computer, enrollment metadata associated with the enrollee-user; and
generating, by the computer, an enrolled metadata embedding for the enrollee-user by applying an identification engine of the machine-learning architecture on a plurality of types of features extracted from the enrollment metadata.
4. The method according to claim 3 , further comprising:
generating, by the computer, a first inbound metadata embedding corresponding to the first inbound user by applying the identification engine of the machine-learning architecture on the plurality of types of features extracted from the enrollment biometric data; and
identifying, by the computer, in the database a set of one or more enrollee-records having corresponding enrolled metadata embeddings having a context similarity score to the first inbound metadata embedding within a context similarity threshold.
5. The method according to claim 1 , wherein the computer selects from the database a set of one or more potential enrollee-records based upon a corresponding context similarity score satisfying a context similarity threshold, the set of one or more potential enrollee-records including the enrollee-record.
6. The method according to claim 1 , further comprising:
applying, by the computer, a risk score engine of the machine-learning architecture on the first inbound contact data of the first authentication request to generate a risk score; and
adjusting, by the computer, at least one of a context similarity threshold or the first biometric-print similarity threshold.
7. The method according to claim 1 , further comprising generating, by the computer, a first result indication for the first provider server based upon the similarity between the first inbound biometric-print and the enrolled biometric-print of the one or more enrolled biometric-prints in the enrollee-record indicating a likelihood that a first inbound audio signal originated from the enrollee-user.
8. The method according to claim 1 , further comprising determining, by the computer, a first similarity score based upon a distance between the first inbound biometric-print and the enrolled biometric-print in a user account for the enrollee-user.
9. The method according to claim 1 , further comprising:
obtaining, by the computer, enrollment data of the enrollee-user for a user account for the enrollee-user, the enrollment data including one or more enrollment audio signals containing one or more utterances of the enrollee-user;
extracting, by the computer, a plurality of enrollment features from the one or more enrollment audio signals for the enrollee-user; and
generating, by the computer, an enrollment voiceprint of the one or more enrollment biometric-prints by applying the machine-learning architecture on the plurality of enrollment features.
10. The method according to claim 1 , further comprising authenticating, by the computer, the first inbound user as the enrollee-user associated with the enrolled biometric-print, based upon the similarity between the first inbound biometric-print and the enrolled biometric-print in the enrollee-record.
11. A system comprising:
a database comprising a non-transitory machine-readable storage configured to store a plurality of enrollee-records for a plurality of enrollee-users; and
a computer comprising a processor and in communication with the database, the computer configured to:
generate one or more enrolled biometric-prints for an enrollee-user in an enrollee-record of the database, by applying one or more biometric recognition engines of a machine-learning architecture on enrollment biometric data for the enrollee-user;
receive a first authentication request from a first provider server to authenticate a first inbound user, and a second authentication request from a second provider server to authenticate a second inbound user;
generate a first inbound biometric-print and a second inbound biometric-print corresponding to the first inbound user and the second inbound user, by applying the one or more biometric engines on corresponding first inbound contact data and second inbound contact data; and
transmit a first result for a first computing service to the first provider server based upon a similarity between the first inbound biometric-print and an enrolled biometric-print of the one or more enrolled biometric-prints in the enrollee-record in accordance with a first biometric-print similarity threshold determined based on the first inbound contact data, and a second result for a second computing service to the second provider server based upon the similarity between the second inbound biometric-print and the enrolled biometric-print of the one or more enrolled biometric-prints in the enrollee-record in accordance with a second biometric-print similarity threshold determined based on the second inbound contact data.
12. The system according to claim 11 , wherein the computer is further configured to update the enrollee-record to indicate one or more computing services associated with the enrollee-user.
13. The system according to claim 11 , wherein the computer is further configured to:
obtain enrollment metadata associated with the enrollee-user; and
generate an enrolled metadata embedding for the enrollee-user by applying an identification engine of the machine-learning architecture on a plurality of types of features extracted from the enrollment metadata.
14. The system according to claim 13 , wherein the computer is further configured to:
generate a first inbound metadata embedding corresponding to the first inbound user by applying the identification engine of the machine-learning architecture on the plurality of types of features from the enrollment biometric data; and
identify in the database a set of one or more enrollee-records having corresponding enrolled metadata embeddings having a context similarity score to the first inbound metadata embedding within a context similarity threshold.
15. The system according to claim 11 , wherein the computer selects from the database a set of one or more potential enrollee-records based upon a corresponding context similarity score satisfying a context similarity threshold, the set of one or more potential enrollee-records including the enrollee-record.
16. The system according to claim 11 , wherein the computer is further configured to:
apply a risk score engine of the machine-learning architecture on the inbound contact data of the first authentication request to generate a risk score; and
adjust at least one of a context similarity threshold or the first biometric-print similarity threshold.
17. The system according to claim 11 , wherein the computer is further configured to generate a first result indication for the first provider server based upon the similarity between the first inbound biometric-print and the enrolled biometric-print of the one or more enrolled biometric-prints in the enrollee-record indicating a likelihood that a first inbound audio signal originated from the enrollee-user.
18. The system according to claim 11 , wherein the computer is further configured to determine a first similarity score based upon a distance between the first inbound biometric-print and the enrolled biometric-print in a user account for the enrollee-user.
19. The system according to claim 11 , wherein the computer is further configured to:
obtain enrollment data of the enrollee-user for a user account for the enrollee-user, the enrollment data including one or more enrollment audio signals containing one or more utterances of the enrollee-user;
extract a plurality of enrollment features from the one or more enrollment audio signals for the enrollee-user; and
generate an enrollment voiceprint of the one or more enrollment biometric-prints by applying the machine-learning architecture on the plurality of enrollment features.
20. The system according to claim 11 , wherein the computer is further configured to authenticate the first inbound user as the enrollee-user associated with the enrolled biometric-print, based upon the similarity between the first inbound biometric-print and the enrolled biometric-print in the enrollee-record.