Method and apparatus with speaker authentication and/or training

Patent metadata

Field	Value
Publication number	US-11367451-B2
Application number	US-201916519757-A
Country	US
Kind code	B2
Filing date	Jul 23, 2019
Priority date	Aug 27, 2018
Publication date	Jun 21, 2022
Grant date	Jun 21, 2022

Abstract

A speaker authentication method and apparatus may extract input speaker features corresponding to a plurality of frames of an input speech of an object, estimate discriminable speaker sections corresponding to the plurality of frames, and dynamically match the input speaker features to pre-enrolled enrolled speaker features based on the discriminable speaker section.

First claim

What is claimed is:

1. A speaker authentication method, comprising:
receiving a plurality of frames corresponding to an input speech;
extracting input speaker features corresponding to the plurality of frames;
estimating discriminable speaker sections corresponding to the plurality of frames;
dynamically matching the input speaker features to pre-enrolled enrolled speaker features based on the discriminable speaker sections; and
performing a speaker authentication based on a result of the dynamic matching,
wherein the dynamic matching comprises:
assigning a first weight to an input speaker feature corresponding to a pre-determined short pause among the input speaker features;
assigning a second weight to an input speaker feature corresponding to a speech among the input speaker features; and
dynamically matching each of the first weight-assigned input speaker feature and the second weight-assigned input speaker feature to the pre-enrolled enrolled speaker features.

2. The method of claim 1 , wherein the dynamic matching comprises:
selecting input speaker features having discriminable speaker sections greater than or equal to a threshold value; and
dynamically matching the selected input speaker features to the pre-enrolled enrolled speaker features.

3. The method of claim 1 , wherein the dynamic matching comprises:
dropping an input speaker feature having a discriminable speaker section less than a threshold value;
dropping a pre-enrolled enrolled speaker feature corresponding to the dropped input speaker feature; and
dynamically matching remaining input speaker features, excluding the dropped input speaker feature, to remaining enrollment speaker features, excluding the dropped pre-enrolled enrollment speaker feature.

4. The method of claim 1 , wherein the dynamic matching comprises:
assigning a weight to input speaker features having discriminable speaker sections being greater than or equal to a threshold value; and
dynamically matching the weight-assigned input speaker features to the pre-enrolled enrolled speaker features.

5. The method of claim 1 , wherein the dynamic matching comprises:
dropping an input speaker feature corresponding to a short pause among the input speaker features; and
dynamically matching remaining input features, excluding the dropped input speaker feature, to the pre-enrolled enrolled speaker features.

6. The method of claim 1 , wherein the dynamic matching comprises:
aligning the pre-enrolled enrolled speaker features representing phonemes identical to phonemes represented by the input speaker features; and
dynamically matching the input speaker features to the aligned pre-enrolled enrolled speaker features.

7. The method of claim 1 , wherein the performing comprises:
outputting a distance corresponding to the input speech by accumulating results of the dynamic matching; and
performing the speaker authentication based on a result of comparing the distance to a threshold value.

8. The method of claim 1 , wherein the extracting comprises extracting the input speaker features based on per-frequency energies of the plurality of frames.

9. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the speaker authentication method of claim 1 .

10. A speaker authentication apparatus, comprising:
a communication interface configured to receive a plurality of frames corresponding to an input speech; and
a processor configured to:
extract input speaker features corresponding to the plurality of frames;
estimate discriminable speaker sections corresponding to the plurality of frames;
dynamically match the input speaker features to pre-enrolled enrolled speaker features based on the discriminable speaker sections; and
perform a speaker authentication based on a result of the dynamic matching,

wherein, for the dynamic matching, the processor is configured to:
assign a first weight to an input speaker feature corresponding to a pre-determined short pause among the input speaker features;
assign a second weight to an input speaker feature corresponding to a speech among the input speaker features; and
dynamically match each of the first weight-assigned input speaker feature and the second weight-assigned input speaker feature to the pre-enrolled enrolled speaker features.

11. The apparatus of claim 10 , wherein the processor is configured to select input speaker features having discriminable speaker sections greater than or equal to a threshold value, and dynamically match the selected input speaker features to the pre-enrolled enrolled speaker features.

12. The apparatus of claim 10 , wherein the processor is configured to drop an input speaker feature having a discriminable speaker section less than a threshold value, and dynamically match remaining input speaker features, excluding the dropped input speaker feature, to the pre-enrolled enrolled speaker features.

13. The apparatus of claim 10 , wherein the processor is configured to assign a weight to input speaker features having discriminable speaker sections greater than or equal to a threshold value, and dynamically match the weight-assigned input speaker features to the pre-enrolled enrolled speaker features.

14. A speaker authentication method, comprising:
extracting input speaker features corresponding to speech frames;
determining discriminable speaker sections in each of the speech frames;
dynamically matching select input speaker features, of the extracted input speaker features, to pre-enrolled enrolled speaker features based on the discriminable speaker sections satisfying a criteria; and
authenticating a speaker based on the dynamically matched input speaker features,
wherein the dynamic matching comprises:
assigning a first weight to an input speaker feature corresponding to a pre-determined short pause among the input speaker features;
assigning a second weight to an input speaker feature corresponding to a speech among the input speaker features; and
dynamically matching each of the first weight-assigned input speaker feature and the second weight-assigned input speaker feature to the pre-enrolled enrolled speaker features.

15. The method of claim 14 , wherein the input speaker features correspond to phonemes and the discriminable speaker sections comprise of voiced sounds.

16. The method of claim 15 , wherein the criteria is satisfied when a discriminable speaker section of the discriminable speaker sections is greater than or equal to a threshold value.

17. The method of claim 15 , wherein the criteria is determined based on comparisons of relative weights applied to the discriminable speaker sections.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Kim Kyuhong
• Kim Insoo
• Lee Dohwan
• Lee Hana

Classifications

• G10L17/02 Primary

  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 title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title