SOUND RECOGNITION METHOD AND SYSTEM

• US 20120226497A1
• Filed: 02/13/2012
• Published: 09/06/2012
• 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
• Granted Patent: US 9,224,388 B2
• Time in Patent Office: Days
• US Class Current: 704/231
• 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 for generating an anti-model of a sound class, the method comprising:

receiving a plurality of candidate sound data;

receiving a reference sound model of the sound class;

determining a plurality of similarity values between the plurality of candidate sound data and the reference sound model; and

generating the anti-model of the sound class based on at least one candidate sound data having the similarity value within a similarity threshold range.

2. The method of claim 1, wherein the plurality of candidate sound data includes at least one of a candidate sound database and a candidate sound model.

3. The method of claim 2, wherein determining the plurality of similarity values includes calculating a likelihood of the candidate sound database matching the reference sound model.

4. The method of claim 2, wherein determining the plurality of similarity values includes calculating a distance between the candidate sound model and the reference sound model.

5. The method of claim 1, further comprising transmitting the anti-model in response to a request from a device.

6. The method of claim 1, further comprising updating the anti-model based on new candidate sound data.

7. The method of claim 1, wherein at least one of the plurality of candidate sound data is received from at least one sound data provider.

8. The method of claim 7, wherein the at least one of the plurality of candidate sound data received from the at least one sound data provider includes an indication of whether the at least one of the plurality of candidate sound data is accessible by a sound data provider other than the at least one sound data provider.

9. The method of claim 1, wherein

the sound class includes a first sound class and a second sound class;

the reference sound model includes a first reference sound model of the first sound class and a second reference sound model of the second sound class; and

the plurality of similarity values includes a plurality of first similarity values between the plurality of candidate sound data and the first reference sound model and a plurality of second similarity values between the plurality of candidate sound data and the second reference sound model, and

wherein generating the anti-model of the sound class includes generating the anti-model of the first sound class and the second sound class based on at least one candidate sound data having at least one of the first similarity value and the second similarity value within the similarity threshold range.

10. A method for providing a sound recognition service to at least one application in a device, the method comprising:

receiving an input sound;

determining a first likelihood of the input sound matching a target sound class of the at least one application based on a sound model of the target sound class;

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

generating a sound detection result indicating whether the input sound corresponds to the target sound class, based on the first likelihood and the second likelihood.

11. The method of claim 10, wherein generating the sound detection result includes determining if a ratio of the first likelihood to the second likelihood is greater than a threshold.

12. The method of claim 10, further comprising providing the sound detection result to the at least one application if the sound detection result indicates that the input sound corresponds to the target sound class.

13. The method of claim 10, wherein the sound detection result includes at least one of the first likelihood and the second likelihood.

14. The method of claim 10, wherein if the at least one application includes a plurality of target sound classes, the anti-model is indicative of a plurality of non-target sound classes including sound classes other than but similar to at least one of the plurality of target sound classes.

15. The method of claim 10, further comprising:

receiving the anti-model via a network from a system configured to provide the sound model and the anti-model.

16. The method of claim 15, wherein receiving the anti-model includes:

checking whether the anti-model is pre-stored in the device; and

if the anti-model is not pre-stored in the device, receiving the anti-model via the network from the system.

17. The method of claim 15, wherein the anti-model is indicative of sound classes other than but similar to a combination of the target sound class and at least one new target sound class.

18. A system for providing at least one of a sound model and an anti-model to be used in sound recognition, the system comprising:

a receiving unit configured to receive a plurality of candidate sound data; and

an anti-model generation unit configured to;

receive a reference sound model of a sound class;

determine a plurality of similarity values between the plurality of candidate sound data and the reference sound model; and

generate the anti-model of the sound class based on at least one candidate sound data having the similarity value within a similarity threshold range.

19. The system of claim 18, wherein the plurality of candidate sound data includes at least one of a candidate sound database and a candidate sound model.

20. The system of claim 18, further comprising a sound model generation unit configured to generate the sound model based on at least one of the plurality of candidate sound data.

21. The system of claim 18, wherein at least one of the plurality of candidate sound data is a candidate sound model, and the similarity value is indicated by a distance between the candidate sound model and the reference sound model.

22. The system of claim 18, wherein at least one of the plurality of candidate sound data is a candidate sound database, and the similarity value is indicated by a likelihood of the candidate sound database matching the reference sound model.

23. The system of claim 18, wherein

the sound class includes a first sound class and a second sound class;

the reference sound model includes a first reference sound model of the first sound class and a second reference sound model of the second sound class; and

the plurality of similarity values includes a plurality of first similarity values between the plurality of candidate sound data and the first reference sound model and a plurality of second similarity values between the plurality of candidate sound data and the second reference sound model, and

wherein the anti-model generation unit is configured to generate the anti-model of the first sound class and the second sound class based on at least one candidate sound data having at least one of the first similarity values and the second similarity values within the similarity threshold range.

24. The system of claim 18, wherein at least one of the plurality of candidate sound data is received from at least one sound data provider.

25. The system of claim 24, wherein the at least one of the plurality of candidate sound data received from the at least one sound data provider includes an indication of whether the at least one of the plurality of candidate sound data is accessible by a sound data provider other than the at least one sound data provider.

26. A device for providing a sound recognition service to at least one application running on the device, the device comprising:

a sound sensing unit configured to capture an input sound;

a storage unit configured to store a sound model and an anti-model of a target sound class; and

a sound detection unit configured to;

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

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

generate a sound detection result indicating whether the input sound corresponds to the target sound class, based on the first likelihood and the second likelihood.

27. The device of claim 26, further comprising a feature extraction unit configured to extract at least one sound feature from the input sound.

28. The device of claim 26, wherein the sound detection unit is further configured to determine if a ratio of the first likelihood to the second likelihood is greater than a threshold.

29. The device of claim 26, wherein the sound detection unit is further configured to provide the sound detection result to the at least one application if the sound detection result indicates that the input sound corresponds to the target sound class.

30. The device of claim 26, wherein if the at least one application has a plurality of target sound classes, the anti-model is indicative of a plurality of non-target sound classes including sound classes other than but similar to at least one of the plurality of target sound classes.

31. The device of claim 26, further comprising a receiving unit configured to receive the anti-model from a system configured to provide the sound model and the anti-model.

32. The device of claim 31, wherein the receiving unit is further configured to:

check whether the anti-model is pre-stored in the device; and

if the anti-model is not pre-stored in the device, receive the anti-model via a network from the system.

33. The device of claim 31, wherein the anti-model is indicative of sound classes other than but similar to a combination of the target sound class and at least one new target sound class.

34. A non-transitory computer-readable medium comprising instructions for generating an anti-model of a sound class, the instructions causing a processor to perform operations of:

receiving a plurality of candidate sound data;

receiving a reference sound model of the sound class;

determining a plurality of similarity values between the plurality of candidate sound data and the reference sound model; and

generating the anti-model of the sound class based on at least one candidate sound data having the similarity value within a similarity threshold range.

35. The medium of claim 34, wherein the plurality of candidate sound data includes at least one of a candidate sound database and a candidate sound model.

36. The medium of claim 34, wherein at least one of the plurality of candidate sound data is received from at least one sound data provider.

37. The medium of claim 36, wherein the at least one of the plurality of candidate sound data received from the at least one sound data provider includes an indication of whether the at least one of the plurality of candidate sound data is accessible by a sound data provider other than the at least one sound data provider.

38. The medium of claim 34, wherein

the sound class includes a first sound class and a second sound class;

the reference sound model includes a first reference sound model of the first sound class and a second reference sound model of the second sound class; and

the plurality of similarity values includes a plurality of first similarity values between the plurality of candidate sound data and the first reference sound model and a plurality of second similarity values between the plurality of candidate sound data and the second reference sound model, and

wherein generating the anti-model of the sound class includes generating the anti-model of the first sound class and the second sound class based on at least one candidate sound data having at least one of the first similarity values and the second similarity values within the similarity threshold range.

39. A non-transitory computer-readable medium comprising instructions for providing a sound recognition service to at least one application in a device, the instructions causing a processor to perform operations of:

receiving an input sound;

determining a first likelihood of the input sound matching a target sound class of at least one application based on a sound model of the target sound class;

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

generating a sound detection result indicating whether the input sound corresponds to the target sound class, based on the first likelihood and the second likelihood.

40. The medium of claim 39, wherein the operations further include receiving the anti-model via a network from a system configured to provide the sound model and the anti-model.

41. The medium of claim 40, wherein receiving the anti-model includes:

checking whether the anti-model is pre-stored in the device; and

if the anti-model is not pre-stored in the device, receiving the anti-model via the network from the system.

42. The medium of claim 40, wherein the anti-model is indicative of sound classes other than but similar to a combination of the target sound class and at least one new target sound class.

43. A system for providing at least one of a sound model and an anti-model to be used in sound recognition, the system comprising:

means for receiving a plurality of candidate sound data; and

means for generating the anti-model based on at least one of the plurality of candidate sound data,

wherein the means for generating the anti-model is configured to;

receive a reference sound model of a sound class;

determine a plurality of similarity values between the plurality of candidate sound data and the reference sound model; and

generate the anti-model of the sound class based on at least one candidate sound data having the similarity value within a similarity threshold range.

44. The system of claim 43, wherein

the sound class includes a first sound class and a second sound class;

the reference sound model includes a first reference sound model of the first sound class and a second reference sound model of the second sound class; and

the plurality of similarity values includes a plurality of first similarity values between the plurality of candidate sound data and the first reference sound model and a plurality of second similarity values between the plurality of candidate sound data and the second reference sound model, and

wherein the means for generating the anti-model is configured to generate the anti-model of the first sound class and the second sound class based on at least one candidate sound data having at least one of the first similarity values and the second similarity values within the similarity threshold range.

45. The system of claim 43, wherein at least one of the plurality of candidate sound data is received from at least one sound data provider.

46. The system of claim 45, wherein the at least one of the plurality of candidate sound data received from the at least one sound data provider includes an indication of whether the at least one of the plurality of candidate sound data is accessible by a sound data provider other than the at least one sound data provider.

47. A device for providing a sound recognition service to at least one application running on the device, the device comprising:

means for capturing an input sound;

means for storing a sound model and an anti-model of a target sound class; and

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

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

generating a sound detection result indicating whether the input sound corresponds to the target sound class, based on the first likelihood and the second likelihood.

48. The device of claim 47, wherein if the at least one application has a plurality of target sound classes, the anti-model is indicative of a plurality of non-target sound classes including sound classes other than but similar to at least one of the plurality of target sound classes.