Granted Patent B2

US 11,783,841 · App. 17/201,987 · Granted Oct 10, 2023

Inventor: Milind Borkar (Plano, TX)

Assignee: ILLUMA LABS INC.

G10L17/16G10L17/04G06F17/16G10L17/02

Abstract

Claims (63)

1. A method for securely authenticating a speaker between a caller device and a first device, comprising:

generating a user account including a user identifier based on at least one metadata element received from a caller device associated with an audio input;

receiving a partial i-vector generated by the first device based on at least the audio input;

generating a full i-vector based on the partial i-vector; and

associating the user account with the full i-vector, wherein generating the full i-vector utilizes an optimized computation including: generating a feature matrix based on a feature vector; generating a Gaussian Mixture Model (GMM) mean matrix based on a plurality of mean vectors associated with a plurality of GMM components; and generating a delta matrix based on the feature matrix and the GMM mean matrix.

2. The method of claim 1 , further comprising:

generating the partial i-vector from an audio frame of the audio input, and a trained T matrix.

3. The method of claim 1 , wherein the optimized computation further comprises:

generating a first multi-dimensional array comprising a plurality of duplicated matrices, each matrix comprised of a plurality of GMM mean vectors;

generating a multi-dimensional feature matrix, comprising a plurality of feature matrices, each feature matrix corresponding to a feature vector of a single audio frame; and

generating a multi-dimensional delta array based on the first multi-dimensional array and the multi-dimensional feature matrix.

4. The method of claim 1 , wherein the optimized computation further comprises:

detecting diagonal matrices; and

performing only those computations that involve diagonal elements.

5. The method of claim 1 , wherein the optimized computation further comprises:

detecting computations in an intermediate result that generate an off-diagonal element of a matrix which is diagonalized; and

eliminating the computation of the intermediate result.

6. The method of claim 1 , wherein the optimized computation further comprises:

detecting a recurring computation;

precomputing the recurring computation; and

storing a precomputed result in a cache of a processor.

7. The method of claim 1 , wherein the optimized computation further comprises:

detecting a computation between a first matrix and a second matrix; and

replacing the detected computation with an element-by-element computation between the first matrix and the second matrix in response to determining that the replacement will result in the same output.

8. The method of claim 1 , wherein the metadata elements are any of: phone number, Automatic Number Identification (ANI), personal identifier number, International Mobile Equipment Identity (IMEI), International Mobile Subscriber Identity (IMSI), end device make, and end device model.

9. The method of claim 1 , further comprising:

generating a user account in response to determining that the full i-vector is not associated with an existing user account.

10. The method of claim 1 , further comprising:

extracting channel features from the audio input; and

generating the full i-vector further based on the extracted channel features; and

associating the extracted channel features with the generated user account.

11. The method of claim 1 , further comprising:

receiving a second partial i-vector and associated metadata;

generating a second full i-vector based on the second partial i-vector; and

generating a match score between the second full i-vector and the full i-vector.

12. The method of claim 11 , further comprising:

generating a notification that a speaker is authenticated, in response to the match score exceeding a first threshold value; and

generating a notification that the speaker is not authenticated, in response to the match score being below a first threshold.

13. A non-transitory computer readable medium having stored thereon instructions for causing a processing circuitry to perform a process, the process comprising:

generating a user account including a user identifier based on at least one metadata element received from a caller device associated with an audio input;

receiving a partial i-vector generated by a first device based on at least the audio input;

generating a full i-vector based on the partial i-vector; and

associating the user account with the full i-vector, wherein generating the full i-vector utilizes an optimized computation including: generating a feature matrix based on a feature vector; generating a Gaussian Mixture Model (GMM) mean matrix based on a plurality of mean vectors associated with a plurality of GMM components; and generating a delta matrix based on the feature matrix and the GMM mean matrix.

14. A system for securely authenticating a speaker between a caller device and a first device, comprising:

a processing circuitry; and

a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to:

generate a user account including a user identifier based on at least one metadata element received from a caller device associated with an audio input;

receive a partial i-vector generated by the first device based on at least the audio input;

generate a full i-vector based on the partial i-vector; and

associate the user account with the full i-vector, wherein generating the full i-vector utilizes an optimized computation including: generating a feature matrix based on a feature vector; generating a Gaussian Mixture Model (GMM) mean matrix based on a plurality of mean vectors associated with a plurality of GMM components; and generating a delta matrix based on the feature matrix and the GMM mean matrix.

15. The system of claim 14 , wherein the memory contains instructions that when executed by the processing circuitry further configure the system to:

extract channel features from the audio input; and

generate the full i-vector further based on the extracted channel features; and

associate the extracted channel features with the generated user account.

16. The system of claim 14 , wherein the memory contains instructions that when executed by the processing circuitry further configure the system to:

extract features into a feature matrix from an incoming audio call;

generate a partial i-vector, wherein the partial i-vector includes a first low-order statistic;

send the partial i-vector to an authentication server; and

receive from the authentication server a match score generated based on a full i-vector and another i-vector being stored on the authentication server, wherein the full i-vector is generated from the partial i-vector.

17. The system of claim 16 , wherein the memory contains instructions that when executed by the processing circuitry further configure the system to:

send metadata associated with a caller device to an authentication server; wherein the match score is further generated based on a match between the sent metadata, and metadata associated with the another i-vector.

18. The system of claim 14 , wherein generating the partial i-vector is performed in response to detecting a voice by a voice activity detector (VAD) of the first device, the voice originating from a caller device.

19. The system of claim 14 , wherein generating a partial i-vector is performed at a first frequency, and generating the full i-vector is performed at a second frequency, which is lower than the first frequency.

Assignments (2)

Continuity (8)