Reverberation compensation for far-field speaker recognition

Patent metadata

Field	Value
Publication number	US-11862176-B2
Application number	US-202117327379-A
Country	US
Kind code	B2
Filing date	May 21, 2021
Priority date	Aug 22, 2016
Publication date	Jan 2, 2024
Grant date	Jan 2, 2024

Abstract

Techniques are provided for reverberation compensation for far-field speaker recognition. A methodology implementing the techniques according to an embodiment includes receiving an authentication audio signal associated with speech of a user and extracting features from the authentication audio signal. The method also includes scoring results of application of one or more speaker models to the extracted features. Each of the speaker models is trained based on a training audio signal processed by a reverberation simulator to simulate selected far-field environmental effects to be associated with that speaker model. The method further includes selecting one of the speaker models, based on the score, and mapping the selected speaker model to a known speaker identification or label that is associated with the user.

First claim

What is claimed is:

1. At least one non-transitory computer readable medium comprising instructions that, when executed, cause at least one processor to at least:
access a source utterance;
artificially create reverberated speech based on a room dimension, a reflection coefficient, and the source utterance; and
train a far-field machine learning model to recognize speech using the artificially created reverberated speech.

2. The at least one non-transitory computer readable medium of claim 1 , wherein the source utterance is a near field utterance.

3. The at least one non-transitory computer readable medium of claim 1 , wherein the instructions, when executed, cause the at least one processor to train the far-field machine learning model using gradient descent.

4. The at least one non-transitory computer readable medium of claim 1 , wherein the instructions, when executed, cause the at least one processor to store the model in a machine readable storage.

5. The at least one non-transitory computer readable medium of claim 1 , wherein the instructions, when executed, cause the at least one processor to:
access spoken audio; and
utilize the model to identify a person based on the spoken audio.

6. The at least one non-transitory computer readable medium of claim 5 , wherein the instructions, when executed, cause the at least one processor to utilize the model to identify the person without removal of reverberation from the spoken audio.

7. An apparatus to perform speaker recognition, the apparatus comprising:
memory;
instructions in the apparatus;
at least one processor to execute the instructions to cause the at least one processor to at least:
access a source utterance;
artificially create reverberated speech based on a room dimension, a reflection coefficient, and the source utterance; and
train a far-field machine learning model to recognize speech using the artificially created reverberated speech.

8. The apparatus of claim 7 , wherein the source utterance is a near field utterance.

9. The apparatus of claim 7 , wherein the processor is to train the far-field machine learning model using gradient descent.

10. The apparatus of claim 7 , wherein the processor is to store the model in a machine readable storage.

11. The apparatus of claim 7 , wherein the processor is to:
access spoken audio; and
utilize the model to identify a person based on the spoken audio.

12. The apparatus of claim 11 , wherein the processor is to utilize the model to identify the person without removal of reverberation from the spoken audio.

13. A method for speaker recognition, the method comprising:
accessing a source utterance;
artificially creating reverberated speech based on a room dimension, a reflection coefficient, and the source utterance; and
training a far-field machine learning model to recognize speech using the artificially created speech.

14. The method of claim 13 , wherein the source utterance is a near field utterance.

15. The method of claim 13 , wherein the training of the far-field machine learning model is performed using gradient descent.

16. The method of claim 13 , further including storing the model in a machine readable storage.

17. The method of claim 13 , further including:
accessing spoken audio; and
utilizing the model to identify a person based on the spoken audio.

18. The method of claim 17 , wherein utilizing the model to identify the person is performed without removal of reverberation from the spoken audio.

19. An apparatus for speaker recognition, the apparatus comprising:
means for accessing a source utterance;
means for creating artificially reverberated speech based on a room dimension, a reflection coefficient, and the source utterance; and
means for training a far-field machine learning model to recognize speech using the artificially created reverberated speech.

20. The apparatus of claim 19 , wherein the source utterance is a near field utterance.

21. The apparatus of claim 19 , wherein the means for training is to train the far-field machine learning model using gradient descent.

Assignees

• Intel Corp

Inventors

• Cilingir Gokcen
• Biswal Narayan

Classifications

• G10L17/04 Primary

  Training, enrolment or model building · CPC title

• G10L17/06

  Decision making techniques; Pattern matching strategies · CPC title

• G10L17/12

  Score normalisation · CPC title

• G10L17/20

  Pattern transformations or operations aimed at increasing system robustness, e.g. against channel noise or different working conditions · CPC title

• G10L21/0208

  Noise filtering · CPC title