Sound recognition method and system

• US 9,224,388 B2
• Filed: 02/13/2012
• Issued: 12/29/2015
• Est. Priority Date: 03/04/2011
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: You, Kisun, Hwang, Kyu Woong, Kim, Taesu
• Primary Examiner(s): Jackson, Jakieda
• Application Number:
  US13/371,966
• Publication Number:
  US 20120226497A1
• Time in Patent Office: 1,415 Days
• Field of Search: 704/231, 704/233, 704/243, 704/226, 704/235, 704/246, 704/255, 704/250, 704/256, 704/256.1, 704/256.2, 704/256.3, 704/256.5, 704/266, 704/E15.027, 704/E15.029, 704/E17.006, 704/E17.007, 704/239, 704/238
• US Class Current: 1/1
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A method for generating an anti-model of a sound class is disclosed. A plurality of candidate sound data is provided for generating the anti-model. A plurality of similarity values between the plurality of candidate sound data and a reference sound model of a sound class is determined. An anti-model of the sound class is generated based on at least one candidate sound data having the similarity value within a similarity threshold range.

Claims

1. A method comprising:

identifying, at a device, a first target sound class based on a first request from a first application executing at the device;

identifying a second target sound class based on a second request from a second application executing at the device;

receiving an input sound;

determining a first likelihood of the input sound matching the first target sound class based on a sound model of the first target sound class;

determining a second likelihood of the input sound matching at least one non-target sound class based on an anti-model;

generating a first sound detection result for the first target sound class; and

generating a second sound detection result for the second target sound class, wherein the first sound detection result indicates whether the input sound matches the first target sound class and the second sound detection result indicates whether the input sound matches the second target sound class, and wherein the first sound detection result is based on the first likelihood and the second likelihood.

2. The method of claim 1, further comprising providing the first sound detection result to the first application based on the first sound detection result indicating that the input sound corresponds to the first target sound class.

3. The method of claim 1, wherein the anti-model is indicative of a plurality of non-target sound classes including sound classes other than a union of the first target sound class and the second target sound class.

4. The method of claim 1, further comprising receiving the anti-model via a network from a server.

5. The method of claim 1, wherein the anti-model is indicative of a plurality of non-target sound classes including sound classes other than the first target sound class, wherein the second sound detection result is based on comparing the input sound to a second anti-model indicative of a second plurality of non-target sound classes including sound classes other than the second target sound class, and wherein the anti-model and the second anti-model are different.

6. The method of claim 1, wherein the first target sound class and the second target sound class are different, and wherein the second sound detection result is determined based on the first sound detection result and without comparing the input sound to the second target sound class.

7. The method of claim 6, wherein when the first sound detection result indicates that the input sound does not match the first target sound class, the second sound detection result indicates that the input sound does not match the second target sound class.

8. A device comprising:

a microphone configured to capture an input sound; and

a memory storing instructions that when executed by a processor cause the processor to;

identify a first target sound class based on a first request from a first application;

identify a second target sound class based on a second request from a second application;

determine a first likelihood of the input sound matching the first target sound class based on a sound model of the first target sound class;

determine a second likelihood of the input sound matching at least one non-target sound class based on an anti-model;

generate a first sound detection result for the first target sound class; and

generate a second sound detection result for the second target sound class, wherein the first sound detection result indicates whether the input sound matches the first target sound class and the second sound detection result indicates whether the input sound matches the second target sound class, and wherein the first sound detection result is based on the first likelihood and the second likelihood.

9. The device of claim 8, wherein the processor is further configured to extract at least one sound feature from the input sound.

10. The device of claim 8, wherein the processor is further configured to determine whether a ratio of the first likelihood to the second likelihood satisfies a threshold value.

11. The device of claim 8, wherein the processor is further configured to provide the first sound detection result to the first application, in response to the first request, when the first sound detection result indicates that the input sound corresponds to the first target sound class.

12. The device of claim 8, wherein the anti-model is based on the one or more target sound classes.

13. The device of claim 8, further comprising a receiver configured to receive the anti-model from a server.

14. A non-transitory computer-readable medium comprising instructions that, when executed by a processor, cause the processor to perform operations including:

identifying a first target sound class based on a first request from a first application executing at a device;

identifying a second target sound class based on a second request from a second application executing at the device;

receiving an input sound;

determining a first likelihood of the input sound matching the first target sound class based on a sound model of the first target sound class;

determining a second likelihood of the input sound matching at least one non-target sound class based on an anti-model;

generating a first sound detection result for the first target sound class; and

generating a second sound detection result for the second target sound class, wherein the first sound detection result indicates whether the input sound matches the first target sound class and the second sound detection result indicates whether the input sound matches the second target sound class, and wherein the first sound detection result is based on the first likelihood and the second likelihood.

15. The non-transitory computer-readable medium of claim 14, wherein the operations further include receiving the anti-model via a network from a server.

16. The non-transitory computer-readable medium of claim 15, wherein, prior to receiving the anti-model, the operations further include:

determining whether the anti-model is stored in a memory accessible to the processor; and

requesting the anti-model via the network from the server based on a determination that the anti-model is not stored in the memory.

17. The non-transitory computer-readable medium of claim 15, wherein the one or more target sound classes includes multiple target sound classes, and wherein the anti-model is indicative of sound classes other than at least two target sound classes.

18. A device comprising:

means for receiving an input sound;

means for identifying a first target sound class based on a first request from a first application executing at a device and for identifying a second target sound class based on a second request from a second application executing at the device;

means for determining a first likelihood of the input sound matching the first target sound class based on a sound model of the first target sound class;

means for determining a second likelihood of the input sound matching at least one non-target sound class based on an anti-model; and

means for generating a first sound detection result for the first target sound class and generating a second sound detection result for the second target sound class, wherein the first sound detection result indicates whether the input sound matches the first target sound class and the second sound detection result indicates whether the input sound matches the second target sound class, and wherein the first sound detection result is based on the first likelihood and the second likelihood.

19. The device of claim 18, wherein when the one or more target sound classes includes more than one target sound class, the anti-model is indicative of a plurality of non-target sound classes including sound classes other than at least one of the more than one target sound classes.