Dynamic threshold for speaker verification

• US 9,384,738 B2
• Filed: 07/25/2014
• Issued: 07/05/2016
• Est. Priority Date: 06/24/2014
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Foerster, Jakob, Casado, Diego Melendo
• Primary Examiner(s): Baker, Charlotte M
• Application Number:
  US14/340,720
• Publication Number:
  US 20150371639A1
• Time in Patent Office: 711 Days
• Field of Search: 704/232, 704/233, 704/235, 704/246, 704/251, 704/275
• US Class Current: 1/1
• CPC Class Codes: G06F 3/167 Audio in a user interface, e.g. using voice commands for navigating, audio feedback
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• 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 17/08 Use of distortion metrics or a particular distance between probe pattern and reference templates
• CPC Class Codes: G10L 17/12 Score normalisation
• CPC Class Codes: G10L 17/14 Use of phonemic categorisation or speech recognition prior to speaker recognition or verification
• 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 17/24 the user being prompted to utter a password or a predefined phrase
• CPC Class Codes: G10L 25/84 for discriminating voice from noise
• CPC Class Codes: H04M 3/385 using speech signals

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for a dynamic threshold for speaker verification are disclosed. In one aspect, a method includes the actions of receiving, for each of multiple utterances of a hotword, a data set including at least a speaker verification confidence score, and environmental context data. The actions further include selecting from among the data sets, a subset of the data sets that are associated with a particular environmental context. The actions further include selecting a particular data set from among the subset of data sets based on one or more selection criteria. The actions further include selecting, as a speaker verification threshold for the particular environmental context, the speaker verification confidence score. The actions further include providing the speaker verification threshold for use in performing speaker verification of utterances that are associated with the particular environmental context.

Claims

1. A computer-implemented method comprising:

receiving, for each of multiple utterances of a hotword, a data set including at least (i) a speaker verification confidence score associated with the utterance, and (ii) environmental context data associated with the utterance;

selecting from among the data sets, a subset of the data sets that are associated with a particular environmental context;

selecting a particular data set from among the subset of data sets based on one or more selection criteria;

selecting, as a speaker verification threshold for the particular environmental context, the speaker verification confidence score included in the particular data set; and

providing, for output to a speaker verification system, the speaker verification threshold for use in performing speaker verification of utterances that are associated with the particular environmental context.

2. The method of claim 1, wherein the environmental context data an amount of noise detected immediately preceding receipt of the utterance.

3. The method of claim 1, wherein the environmental context data specifies a loudness of the utterance.

4. The method of claim 1, wherein the environmental context data specifies a signal-to-noise ratio of a loudness of an audio signal that encodes the utterance.

5. The method of claim 1, wherein the one or more selection criteria is an empirically defined rejection rate.

6. The method of claim 1, comprising:

labeling the data sets with a post trigger accuracy indicator associated with the utterance.

7. The method of claim 1, comprising:

labeling the data sets with different, second speaker verification confidence score.

8. The method of claim 1, wherein the data sets each further includes an audio signal that encodes the utterance.

9. The method claim 1, wherein selecting from among the data sets, a subset of the data sets that are associated with a particular environmental context comprises:

determining an environmental context data range; and

selecting the subset of the data sets that includes the environmental context data associated with the utterance within the environmental context data range.

10. The method of claim 1, wherein selecting a particular data set from among the subset of data sets based on one or more selection criteria comprises:

determining a threshold based on the one or more selection criteria; and

identifying the particular data set from among the subset of data sets that satisfies the threshold by less than other data sets in the subset of data sets.

11. The method of claim 1, comprising:

selecting from among the data sets, a plurality of subsets of the data sets that are each associated with a respective particular environmental context;

selecting, based on the one or more selection criteria, a plurality of particular data sets, each particular data set being from among a respective subset of the data sets;

selecting, as a plurality of speaker verification thresholds, each of the speaker verification threshold being for the respective particular environmental context, a plurality of speaker verification confidence scores included in each particular data set; and

providing, for output to the speaker verification system, the plurality of speaker verification thresholds for use in performing speaker verification of utterances that are associated with the respective particular environmental context.

12. The method of claim 1, comprising:

selecting from among the data sets, a plurality of subsets of the data sets that are each associated with a respective user;

selecting, based on the one or more selection criteria, a plurality of particular data sets, each particular data set being from among a respective subset of the data sets;

selecting, as a plurality of speaker verification thresholds, each of the speaker verification threshold being for the respective user, a plurality of speaker verification confidence scores included in each particular data set; and

providing, for output to the speaker verification system, the plurality of speaker verification thresholds for use in performing speaker verification of utterances that are associated with the respective user.

13. The method of claim 1, wherein providing, for output to a speaker verification system, the speaker verification threshold for use in performing speaker verification of utterances that are associated with the particular environmental context comprises:

providing, for output to a user device, an environmental context data range and a speaker verification threshold for the environmental context data range.

14. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising;

receiving, for each of multiple utterances of a hotword, a data set including at least (i) a speaker verification confidence score associated with the utterance, and (ii) environmental context data associated with the utterance;

selecting from among the data sets, a subset of the data sets that are associated with a particular environmental context;

selecting a particular data set from among the subset of data sets based on one or more selection criteria;

selecting, as a speaker verification threshold for the particular environmental context, the speaker verification confidence score included in the particular data set; and

providing, for output to a speaker verification system, the speaker verification threshold for use in performing speaker verification of utterances that are associated with the particular environmental context.

15. The system claim 14, wherein selecting from among the data sets, a subset of the data sets that are associated with a particular environmental context comprises:

determining an environmental context data range; and

selecting the subset of the data sets that includes the environmental context data associated with the utterance within the environmental context data range.

16. The system of claim 14, wherein selecting a particular data set from among the subset of data sets based on one or more selection criteria comprises:

determining a threshold based on the one or more selection criteria; and

identifying the particular data set from among the subset of data sets that satisfies the threshold by less than other data sets in the subset of data sets.

17. The system of claim 14, wherein the operations further comprise:

selecting from among the data sets, a plurality of subsets of the data sets that are each associated with a respective particular environmental context;

selecting, based on the one or more selection criteria, a plurality of particular data sets, each particular data set being from among a respective subset of the data sets;

selecting, as a plurality of speaker verification thresholds, each of the speaker verification threshold being for the respective particular environmental context, a plurality of speaker verification confidence scores included in each particular data set; and

providing, for output to the speaker verification system, the plurality of speaker verification thresholds for use in performing speaker verification of utterances that are associated with the respective particular environmental context.

18. The system of claim 14, wherein the operations further comprise:

selecting from among the data sets, a plurality of subsets of the data sets that are each associated with a respective user;

selecting, based on the one or more selection criteria, a plurality of particular data sets, each particular data set being from among a respective subset of the data sets;

selecting, as a plurality of speaker verification thresholds, each of the speaker verification threshold being for the respective user, a plurality of speaker verification confidence scores included in each particular data set; and

providing, for output to the speaker verification system, the plurality of speaker verification thresholds for use in performing speaker verification of utterances that are associated with the respective user.

19. The system of claim 14, wherein providing, for output to a speaker verification system, the speaker verification threshold for use in performing speaker verification of utterances that are associated with the particular environmental context comprises:

providing, for output to a user device, an environmental context data range and a speaker verification threshold for the environmental context data range.

20. A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

receiving, for each of multiple utterances of a hotword, a data set including at least (i) a speaker verification confidence score associated with the utterance, and (ii) environmental context data associated with the utterance;

selecting from among the data sets, a subset of the data sets that are associated with a particular environmental context;

selecting a particular data set from among the subset of data sets based on one or more selection criteria;

selecting, as a speaker verification threshold for the particular environmental context, the speaker verification confidence score included in the particular data set; and

providing, for output to a speaker verification system, the speaker verification threshold for use in performing speaker verification of utterances that are associated with the particular environmental context.