Speaker recognition using adaptive thresholding

• US 9,972,322 B2
• Filed: 03/29/2016
• Issued: 05/15/2018
• Est. Priority Date: 03/29/2016
• Status: Active Grant

• Current Assignee: Hyundai Motor Company (Hyundai Motor Group), Kia Corp.
• Original Assignee: Intel Corporation
• Inventors: Biswal, Narayan, Cilingir, Gokcen
• Primary Examiner(s): Hang, Vu B
• Application Number:
  US15/083,548
• Publication Number:
  US 20170287490A1
• Time in Patent Office: 777 Days
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 17/02 Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/20 Pattern transformations or operations aimed at increasing system robustness, e.g. against channel noise or different working conditions
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces
• CPC Class Codes: G10L 25/78 Detection of presence or absence of voice signals switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10
• CPC Class Codes: G10L 25/81 for discriminating voice from music

Abstract

Techniques related to speaker recognition are discussed. Such techniques may include determining an adaptive speaker recognition threshold based on a speech to noise ratio and noise type label corresponding to received audio and performing speaker recognition based on the adaptive speaker recognition threshold and a speaker recognition score corresponding to received audio.

Claims

1. A computer-implemented method for speaker recognition comprising:

determining a speaker recognition score based on a received audio input;

generating a speech to noise ratio based on the received audio input;

generating a noise type label corresponding to the received audio input;

determining a selected adaptive speaker recognition threshold from a plurality of adaptive speaker recognition thresholds based on the speech to noise ratio, the noise type label, and a target false accept rate corresponding to the received audio input, wherein each of the plurality of adaptive speaker recognition thresholds is associated with a particular noise type label, a particular speech to noise ratio, and a particular target false accept rate, and wherein each of the plurality of adaptive speaker recognition thresholds provides a threshold corresponding to a lowest target false reject rate for the particular noise type label, the particular speech to noise ratio, and the particular target false accept rate; and

performing speaker recognition for the received audio input based on a comparison of the speaker recognition score to the selected adaptive speaker recognition threshold.

2. The method of claim 1, further comprising:

determining the target false accept rate based on a speaker recognition application corresponding to the received audio input.

3. The method of claim 2, wherein the target false accept rate comprises a first value when the speaker recognition application comprises a biometric security application and a second value when the speaker recognition application comprises an authenticated command application, and wherein the first value is less than the second value.

4. The method of claim 1, wherein determining the selected adaptive speaker recognition threshold comprises accessing a look up table based on the speech to noise ratio, the noise type label, and the target false accept rate.

5. The method of claim 1, further comprising:

determining a second selected adaptive speaker recognition threshold from the plurality of adaptive speaker recognition thresholds based on a second speech to noise ratio, a second noise type label, and a second target false accept rate corresponding to a second received audio input; and

performing a second speaker recognition for the second received audio input based on a second comparison of a second speaker recognition score to the second selected adaptive speaker recognition threshold, wherein the first and second selected adaptive speaker recognition thresholds are different.

6. The method of claim 1, wherein performing speaker recognition comprises accepting the received audio input as corresponding to a target user when the speaker recognition score exceeds the selected adaptive speaker recognition threshold or rejecting the received audio input as corresponding to the target user when the speaker recognition score does not exceed the selected adaptive speaker recognition threshold.

7. The method of claim 1, further comprising:

determining a second speaker recognition score based on the received audio input, wherein determining the speaker recognition score comprises applying a first speaker model corresponding to a first user and determining the second speaker recognition score comprises applying a second speaker model corresponding to a second user, wherein the first and second speaker models are different, and

wherein performing the speaker recognition comprises a second comparison of the second speaker recognition score to the selected adaptive speaker recognition threshold.

8. The method of claim 7, wherein the speaker recognition score and the second speaker recognition score both exceed the selected adaptive speaker recognition threshold, the method further comprising:

providing an identified speaker indicator corresponding to the first user based on the speaker recognition score exceeding the second speaker recognition score.

9. The method of claim 1, wherein at least one of determining the speaker recognition score, generating the speech to noise ratio, or generating the noise type label comprises extracting features from the received audio input and applying a pretrained classification model to the features.

10. A system for performing speaker recognition comprising:

a memory configured to store a received audio input; and

a digital signal processor coupled to the memory, the digital signal processor to determine a speaker recognition score based on a received audio input, to generate a speech to noise ratio based on the received audio input, to generate a noise type label corresponding to the received audio input, to determine a selected adaptive speaker recognition threshold from a plurality of adaptive speaker recognition thresholds based on the speech to noise ratio, the noise type label, and a target false accept rate corresponding to the received audio input, wherein each of the plurality of adaptive speaker recognition thresholds is associated with a particular noise type label, a particular speech to noise ratio, and a particular target false accept rate, and wherein each of the plurality of adaptive speaker recognition thresholds provides a threshold corresponding to a lowest target false reject rate for the particular noise type label, the particular speech to noise ratio, and the particular target false accept rate, and to perform speaker recognition for the received audio input based on a comparison of the speaker recognition score to the selected adaptive speaker recognition threshold.

11. The system of claim 10, wherein the digital signal processor is to determine the target false accept rate based on a speaker recognition application corresponding to the received audio input.

12. The system of claim 10, wherein the digital signal processor to determine the adaptive speaker recognition threshold comprises the digital signal processor to access a look up table based on the speech to noise ratio, the noise type label, and the target false accept rate.

13. The system of claim 10, wherein the digital signal processor is to determine a second selected adaptive speaker recognition threshold from the plurality of adaptive speaker recognition thresholds based on a second speech to noise ratio, a second noise type label, and a second target false accept rate corresponding to a second received audio input and to perform a second speaker recognition for the second received audio input based on a second comparison of a second speaker recognition score to the second selected adaptive speaker recognition threshold, wherein the first and second selected adaptive speaker recognition thresholds are different.

14. The system of claim 10, wherein the digital signal processor to perform speaker recognition comprises the digital signal processor to accept the received audio input as corresponding to a target user when the speaker recognition score exceeds the selected adaptive speaker recognition threshold or to reject the received audio input as corresponding to the target user when the speaker recognition score does not exceed the selected adaptive speaker recognition threshold.

15. The system of claim 10, wherein the digital signal processor is to determine a second speaker recognition score based on the received audio input, wherein to determine the speaker recognition score comprises the digital signal processor to apply a first speaker model corresponding to a first user and to determine the second speaker recognition score comprises the digital signal processor to apply a second speaker model corresponding to a second user, wherein the first and second speaker models are different, and

wherein the digital signal processor to perform the speaker recognition comprises the digital signal processor to perform a second comparison of the second speaker recognition score to the selected adaptive speaker recognition threshold.

23. The system of claim 14, wherein the speaker recognition score and the second speaker recognition score both exceed the selected adaptive speaker recognition threshold and the digital signal processor is further to provide an identified speaker indicator corresponding to the first user based on the speaker recognition score exceeding the second speaker recognition score.

16. At least one non-transitory machine readable medium comprising a plurality of instructions that, in response to being executed on a device, cause the device to provide speaker recognition by:

determining a speaker recognition score based on a received audio input;

generating a speech to noise ratio based on the received audio input;

generating a noise type label corresponding to the received audio input;

determining a selected adaptive speaker recognition threshold from a plurality of adaptive speaker recognition thresholds based on the speech to noise ratio, the noise type label, and a target false accept rate corresponding to the received audio input, wherein each of the plurality of adaptive speaker recognition thresholds is associated with a particular noise type label, a particular speech to noise ratio, and a particular target false accept rate, and wherein each of the plurality of adaptive speaker recognition thresholds provides a threshold corresponding to a lowest target false reject rate for the particular noise type label, the particular speech to noise ratio, and the particular target false accept rate; and

performing speaker recognition for the received audio input based on a comparison of the speaker recognition score to the selected adaptive speaker recognition threshold.

17. The machine readable medium of claim 16, further comprising a plurality of instructions that, in response to being executed on the device, cause the device to provide speaker recognition by:

determining the target false accept rate based on a speaker recognition application corresponding to the received audio input.

18. The machine readable medium of claim 16, wherein determining the selected adaptive speaker recognition threshold comprises accessing a look up table based on the speech to noise ratio, the noise type label, and the target false accept rate.

19. The machine readable medium of claim 16, further comprising a plurality of instructions that, in response to being executed on the device, cause the device to provide speaker recognition by:

determining a second selected adaptive speaker recognition threshold from the plurality of adaptive speaker recognition thresholds based on a second speech to noise ratio, a second noise type label, and a second target false accept rate corresponding to a second received audio input; and

performing a second speaker recognition for the second received audio input based on a second comparison of a second speaker recognition score to the second selected adaptive speaker recognition threshold, wherein the first and second selected adaptive speaker recognition thresholds are different.

20. The machine readable medium of claim 16, wherein performing speaker recognition comprises accepting the received audio input as corresponding to a target user when the speaker recognition score exceeds the selected adaptive speaker recognition threshold or rejecting the received audio input as corresponding to the target user when the speaker recognition score does not exceed the selected adaptive speaker recognition threshold.

21. The machine readable medium of claim 16, further comprising a plurality of instructions that, in response to being executed on the device, cause the device to provide speaker recognition by:

determining a second speaker recognition score based on the received audio input, wherein determining the speaker recognition score comprises applying a first speaker model corresponding to a first user and determining the second speaker recognition score comprises applying a second speaker model corresponding to a second user, wherein the first and second speaker models are different, and

wherein performing the speaker recognition comprises a second comparison of the second speaker recognition score to the selected adaptive speaker recognition threshold.

22. The machine readable medium of claim 21, wherein the speaker recognition score and the second speaker recognition score both exceed the selected adaptive speaker recognition threshold, the machine readable medium further comprising a plurality of instructions that, in response to being executed on the device, cause the device to provide speaker recognition by:

providing an identified speaker indicator corresponding to the first user based on the speaker recognition score exceeding the second speaker recognition score.