WEARER VOICE ACTIVITY DETECTION

• US 20170178668A1
• Filed: 12/22/2015
• Published: 06/22/2017
• Est. Priority Date: 12/22/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Intel Corporation
• Inventors: Kar, Swarnendu, Phatak, Gauri, Shah, Saurin
• Granted Patent: US 9,978,397 B2
• Time in Patent Office: Days
• 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 17/18 Artificial neural networks; Connectionist approaches
• CPC Class Codes: G10L 21/0232 Processing in the frequency domain
• 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

Abstract

Embodiments include a wearable device, such as a head-worn device. The wearable device includes a first microphone to receive a first sound signal from a wearer of the wearable device; a second microphone to receive a second sound signal from the wearer of the wearable device; and a processor to process the first sound signals and the second sound signals to determine that the first and second sound signals originate from the wearer of the wearable device.

Claims

1. A wearable device comprising:

a first microphone to receive a first sound signal from a wearer of the wearable device;

a second microphone to receive a second sound signal from the wearer of the wearable device;

a processor to process the first sound signals and the second sound signals to determine that the first and second sound signals originate from the wearer of the wearable device.

2. The wearable device of claim 1, wherein the first microphone comprises an air microphone.

3. The wearable device of any of claim 1, wherein the second microphone comprises one of a bone conduction microphone or an in-ear microphone.

4. The wearable device of claim 1, wherein the processor is configured to process the first sound signal and the second sound signal by:

sampling the first sound signal and the second sound signal;

extracting one or more features from the first sound signal and from the second sound signal; and

determining that the first and second sound signals originate from the wearer by comparing extracted features from the first sound signal and from the second sound signal.

5. The wearable device of any of claim 1, further comprising a neural network to process extracted features from each of the first sound signal and from the second sound signal.

6. The wearable device of claim 5, wherein the neural network is trained with a voice dataset and a background dataset, and wherein the neural network is configured to determine based on the voice dataset, the background dataset, and the extracted features from the first and second sound signals that the first and second sound signals original from the wearer of the wearable device.

7. The wearable device of any of claim 1, further comprising a Fast Fourier Transform module to filter the first and second sound signals.

8. The wearable device of any of claim 1, wherein the processor is configured to process the first sound signal and the second sound signal by:

splitting the first sound signal into a first set of subparts, the first set of subparts comprising a frame representing a portion in time of the first sound signal;

splitting the second sound signal into a second set of subparts, the second set of subparts comprising a frame representing a portion in time of the second sound signal;

combining a frame from the first set of subparts with a corresponding frame from the second set of subparts; and

extracting a feature of the first and second sound signals based on combining the frame from the first set of subparts with the corresponding frame from the second set of subparts.

9. The wearable device of claim 1, wherein the wearable device comprises a head-worn device.

10. A method comprising:

receiving a first sound signal at a wearable device from a first microphone;

receiving a second sound signal at the wearable device from a second microphone;

determining that the first and second sound signals originate from the wearer of the wearable device.

11. The method of claim 10, wherein the first microphone comprises an air microphone.

12. The method of any of claim 10, wherein the second microphone comprises one of a bone conduction microphone or an in-ear microphone.

13. The method of claim 10, wherein processing the first sound signal and the second sound signal comprises

sampling the first sound signal and the second sound signal;

extracting one or more features from the first sound signal and from the second sound signal; and

determining that the first and second sound signals originate from the wearer by comparing extracted features form the first sound signal and from the second sound signal.

14. The method of any of claim 10, further comprising processing the extracted features from each of the first sound signal and from the second sound signal using a neural network.

15. The method of claim 14, wherein the neural network is trained with a voice dataset and a background dataset, and wherein the neural network is configured to determine based on the voice dataset, the background dataset, and the extracted features from the first and second sound signals that the first and second sound signals original from the wearer of the wearable device.

16. The method of any of claim 10, further comprising filtering the first and second sound signals using a Fast Fourier Transform module.

17. The method of any of claim 10, wherein processing the first sound signal and the second sound signal comprises:

splitting the first sound signal into a first set of subparts, the first set of subparts comprising a frame representing a portion in time of the first sound signal;

splitting the second sound signal into a second set of subparts, the second set of subparts comprising a frame representing a portion in time of the second sound signal;

combining a frame from the first set of subparts with a corresponding frame from the second set of subparts; and

extracting a feature of the first and second sound signals based on combining the frame from the first set of subparts with the corresponding frame from the second set of subparts.

18. A computer program product tangibly embodied on a non-transient computer readable medium, the computer program product comprising instructions operable when executed to:

receive from a first microphone a first sound signal from a wearer of the wearable device;

receive from a second microphone a second sound signal from the wearer of the wearable device; and

process the first sound signals and the second sound signals to determine that the first and second sound signals originate from the wearer of the wearable device.

19. The computer program product of claim 18, wherein the first microphone comprises an air microphone.

20. The computer program product of claim 18, wherein the second microphone comprises one of a bone conduction microphone or an in-ear microphone.

21. The computer program product of claim 18, further comprising instructions operable when executed to:

sample the first sound signal and the second sound signal;

extract one or more features from the first sound signal and from the second sound signal; and

determine that the first and second sound signals originate from the wearer by comparing extracted features form the first sound signal and from the second sound signal.

22. The computer program product of claim 18, further comprising a neural network to process extracted features from each of the first sound signal and from the second sound signal.

23. The computer program product of claim 22, wherein the neural network is trained with a voice dataset and a background dataset, and wherein the neural network is configured to determine based on the voice dataset, the background dataset, and the extracted features from the first and second sound signals that the first and second sound signals original from the wearer of the wearable device.

24. The computer program product of claim 18, further comprising instructions operable when executed to filter the first and second sound signals based on a Fast Fourier Transform.

25. The computer program product of claim 18, further comprising instructions operable when executed to:

splitting the first sound signal into a first set of subparts, the first set of subparts comprising a frame representing a portion in time of the first sound signal;

splitting the second sound signal into a second set of subparts, the second set of subparts comprising a frame representing a portion in time of the second sound signal;

combining a frame from the first set of subparts with a corresponding frame from the second set of subparts; and

extracting a feature of the first and second sound signals based on combining the frame from the first set of subparts with the corresponding frame from the second set of subparts.