Granted Patent B2

US 12,451,135 · App. 18/297,509 · Granted Oct 21, 2025

Inventors: Cindy Sushen Tseng (Santa Clara, CA); Srinivasa Rao Ponakala (Sunnyvale, CA); Myungjong Kim (Milpitas, CA); Taeyeon Ki (Milpitas, CA); Vijendra Raj Apsingekar (San Jose, CA)

Assignee: Samsung Electronics Co., Ltd.

G10L15/22G10L15/08G10L25/51G10L2015/088G10L2015/223

Abstract

Claims (86)

1. A method comprising:

obtaining, using at least one processing device of an electronic device, an audio input and a location, the location associated with the electronic device;

generating, using the at least one processing device, an audio embedding associated with the audio input;

determining, using the at least one processing device, a first difference between the audio embedding associated with the audio input and an audio embedding associated with a known user;

determining, using the at least one processing device, a second difference between the location associated with the electronic device and a known location associated with the known user;

generating, using the at least one processing device and a false trigger mitigation (FTM) system, a probability of the audio input including a false trigger for automatic speech recognition based on the audio input, the first difference, and the second difference; and

determining, using the at least one processing device, whether to perform automatic speech recognition based on the probability.

2. The method of claim 1 , wherein the FTM system is configured to compare an utterance length to a threshold length and increase the probability of the audio input including the false trigger based on how much the utterance length exceeds the threshold length, the utterance length indicative of a duration of speech contained in the audio input.

3. The method of claim 1 , wherein the FTM system is configured to increase the probability of the audio input including the false trigger based on the first difference such that larger differences between the audio embedding associated with the audio input and the audio embedding associated with the known user increase the probability of the audio input including the false trigger.

4. The method of claim 3 , wherein the FTM system is configured to increase the probability of the audio input including the false trigger based on the second difference such that larger differences between the location associated with the electronic device and the known location associated with the known user increase the probability of the audio input including the false trigger.

5. The method of claim 4 , wherein:

the known user represents an enrolled user associated with the electronic device;

the audio embedding associated with the known user represents an audio embedding generated during or after enrollment of the enrolled user; and

the known location associated with the known user represents a location of a home of the enrolled user.

6. The method of claim 1 , further comprising:

initiating automatic speech recognition of the audio input in response to determining that the probability of the audio input including the false trigger is below a threshold probability; and

performing or initiating performance of one or more actions based on a result of the automatic speech recognition.

7. The method of claim 1 , wherein the FTM system generates the probability of the audio input including the false trigger using a formula of:

p ( Y′|X )= p ( Y|X )*(α 0 max(0,|Y|−T)/β 0 +α 1 |U embed −E embed |/β 1 +α 2 |U loc −E loc |/β 2 )

where:

p(Y′|X) represents the probability of the audio input including the false trigger;

p(Y|X) represents an initial probability of the audio input including the false trigger;

α 0 max(0,|Y|−T)/β 0 represents a penalty associated with an utterance length indicative of a duration of speech contained in the audio input;

α 1 |U embed −E embed |/β 1 represents a penalty associated with the first difference;

α 2 |U loc −E loc |/β 2 represents a penalty associated with the second difference;

|Y| represents the utterance length;

T represents a threshold length;

α 0 and β 0 represent tunable parameters associated with the utterance length;

U embed represents the audio embedding associated with the audio input;

E embed represents the audio embedding associated with the known user;

α 1 and β 1 represent tunable parameters associated with the first difference;

U loc represents the location associated with the electronic device;

E loc represents the known location associated with the known user; and

α 2 and β 2 represent tunable parameters associated with the second difference.

8. An electronic device comprising:

at least one processing device configured to:

obtain an audio input and a location, the location associated with the electronic device;

generate an audio embedding associated with the audio input;

determine a first difference between the audio embedding associated with the audio input and an audio embedding associated with a known user;

determine a second difference between the location associated with the electronic device and a known location associated with the known user;

generate, using a false trigger mitigation (FTM) system, a probability of the audio input including a false trigger for automatic speech recognition based on the audio input, the first difference, and the second difference; and

determine whether to perform automatic speech recognition based on the probability.

9. The electronic device of claim 8 , wherein the FTM system is configured to compare an utterance length to a threshold length and increase the probability of the audio input including the false trigger based on how much the utterance length exceeds the threshold length, the utterance length indicative of a duration of speech contained in the audio input.

10. The electronic device of claim 8 , wherein the FTM system is configured to increase the probability of the audio input including the false trigger based on the first difference such that larger differences between the audio embedding associated with the audio input and the audio embedding associated with the known user increase the probability of the audio input including the false trigger.

11. The electronic device of claim 10 , wherein the FTM system is configured to increase the probability of the audio input including the false trigger based on the second difference such that larger differences between the location associated with the electronic device and the known location associated with the known user increase the probability of the audio input including the false trigger.

12. The electronic device of claim 11 , wherein:

the known user represents an enrolled user associated with the electronic device;

the audio embedding associated with the known user represents an audio embedding generated during or after enrollment of the enrolled user; and

the known location associated with the known user represents a location of a home of the enrolled user.

13. The electronic device of claim 8 , wherein the at least one processing device is further configured to:

initiate automatic speech recognition of the audio input in response to determining that the probability of the audio input including the false trigger is below a threshold probability; and

perform or initiate performance of one or more actions based on a result of the automatic speech recognition.

14. The electronic device of claim 8 , wherein the FTM system is configured to generate the probability of the audio input including the false trigger using a formula of:

p ( Y′|X )= p ( Y|X )*(α 0 max(0,|Y|−T)/β 0 +α 1 |U embed −E embed |/β 1 +α 2 |U loc −E loc |/β 2 )

where:

p(Y′|X) represents the probability of the audio input including the false trigger;

p(Y|X) represents an initial probability of the audio input including the false trigger;

α 0 max(0,|Y|−T)/β 0 represents a penalty associated with an utterance length indicative of a duration of speech contained in the audio input;

α 1 |U embed −E embed |/β 1 represents a penalty associated with the first difference;

α 2 |U loc −E loc |/β 2 represents a penalty associated with the second difference;

|Y| represents the utterance length;

T represents a threshold length;

α 0 and β 0 represent tunable parameters associated with the utterance length;

U embed represents the audio embedding associated with the audio input;

E embed represents the audio embedding associated with the known user;

α 1 and β 1 represent tunable parameters associated with the first difference;

U loc represents the location associated with the electronic device;

E loc represents the known location associated with the known user; and

α 2 and β 2 represent tunable parameters associated with the second difference.

15. A non-transitory machine readable medium containing instructions that when executed cause at least one processor of an electronic device to:

obtain an audio input and a location, the location associated with the electronic device;

generate an audio embedding associated with the audio input;

determine a first difference between the audio embedding associated with the audio input and an audio embedding associated with a known user;

determine a second difference between the location associated with the electronic device and a known location associated with the known user;

generate, using a false trigger mitigation (FTM) system, a probability of the audio input including a false trigger for automatic speech recognition based on the audio input, the first difference, and the second difference; and

determine whether to perform automatic speech recognition based on the probability.

16. The non-transitory machine readable medium of claim 15 , wherein the FTM system is configured to compare an utterance length to a threshold length and increase the probability of the audio input including the false trigger based on how much the utterance length exceeds the threshold length, the utterance length indicative of a duration of speech contained in the audio input.

17. The non-transitory machine readable medium of claim 15 , wherein the FTM system is configured to increase the probability of the audio input including the false trigger based on the first difference such that larger differences between the audio embedding associated with the audio input and the audio embedding associated with the known user increase the probability of the audio input including the false trigger.

18. The non-transitory machine readable medium of claim 17 , wherein the FTM system is configured to increase the probability of the audio input including the false trigger based on the second difference such that larger differences between the location associated with the electronic device and the known location associated with the known user increase the probability of the audio input including the false trigger.

19. The non-transitory machine readable medium of claim 18 , wherein:

the known user represents an enrolled user associated with the electronic device;

the audio embedding associated with the known user represents an audio embedding generated during or after enrollment of the enrolled user; and

the known location associated with the known user represents a location of a home of the enrolled user.

20. The non-transitory machine readable medium of claim 15 , further containing instructions that when executed cause the at least one processor to:

initiate automatic speech recognition of the audio input in response to determining that the probability of the audio input including the false trigger is below a threshold probability; and

perform or initiate performance of one or more actions based on a result of the automatic speech recognition.

Assignments (1)

Continuity (2)

References Cited (76)