Granted Patent B1

US 12,494,200 · App. 18/082,742 · Granted Dec 9, 2025

Inventors: Ahmet Emre Barut (New York, NY); Melanie C B Gens (Honolulu, HI); Matthew Cavell Johnson (Maple Valley, WA); Prashan Wanigasekara (Arlington, MA); Chengwei Su (Lexington, MA); Kechin Qin (Medford, MA); Fan Yang (Waltham, MA); Spurthideepika Sandiri (Holliston, MA)

Assignee: Amazon Technologies, Inc.

G10L15/22G06F3/167G10L15/1815G10L15/19G10L15/24G10L25/57G10L2015/223G10L15/30

Abstract

Claims (100)

1 . A computer-implemented method comprising:

causing a first image to be displayed;

storing first data associating the first image with a first device identifier of a first device;

after causing the first image to be displayed, receiving, from the first device, first input audio data representing a first spoken input;

performing automatic speech recognition (ASR) processing using the first input audio data to generate ASR output data including a first transcript of the first spoken input;

receiving, from the first device, second input audio data representing a second spoken input;

performing ASR processing using the second input audio data to generate second ASR output data including a second transcript of the second spoken input;

based on the first data, the first input audio data, and the second input audio data being received from the first device, determining first embedding data associated with the first image, the first embedding data representing at least one feature of the first image;

processing, using an encoder, the second ASR output data to generate second embedding data representing at least one feature of the second spoken input;

using a first machine learning (ML) component, processing the ASR output data and the first embedding data to determine a first similarity between the ASR output data and the first embedding data, wherein the first similarity represents a likelihood that the first spoken input is requesting performance of an action with respect to the first image;

determining, based on the first similarity, that the first spoken input relates to the first image;

based on determining that the first spoken input relates to the first image, performing a first action responsive to the first spoken input;

using a second ML component, processing the first embedding data and the second embedding data to determine that the second spoken input is associated with the first image; and

based on determining that the second spoken input is associated with the first image, performing a second action responsive to the second spoken input.

2 . The computer-implemented method of claim 1 , further comprising:

prior to receiving the first input audio data, receiving, from the first device, the second input audio data representing the second spoken input;

performing natural language understanding (NLU) processing using the second ASR output data to generate NLU output data including at least an intent corresponding to the second spoken input;

based on the NLU output data, determining the first image, the first image being responsive to the second spoken input; and

ceasing display of the first image,

wherein the first input audio data is received after ceasing display of the first image.

3 . The computer-implemented method of claim 1 , further comprising:

using the second ML component, processing the first embedding data and the second embedding data to determine a second similarity between the first embedding data and the second embedding data,

wherein the second similarity represents a likelihood that the second embedding data is requesting performance of a second action with respect to the first image.

4 . The computer-implemented method of claim 1 , further comprising:

prior to receiving the first input audio data, causing a second image to be displayed, wherein the first data is further stored to associate the second image with the first device identifier;

based on the first data and the first input audio data being received from the first device, determining third embedding data associated with the second image, wherein the third embedding data represents at least one feature of the second image;

using the first ML component, processing the third embedding data to determine a third similarity between the ASR output data and the third embedding data; and

determining, based on the first similarity and the third similarity, that the first spoken input relates to the first image instead of the second image.

5 . A computer-implemented method comprising:

receiving a first representation of a first natural language user input;

determining first content data that was displayed prior to receiving the first representation of the first natural language user input;

determining first embedding data associated with the first content data, the first embedding data representing at least one feature of the first content data;

using the first embedding data, determining that the first natural language user input refers to the first content data;

based on determining that the first natural language user input refers to the first content data, performing a first action responsive to the first natural language user input;

receiving a second representation of a second natural language user input;

determining second embedding data representing at least one feature of the second natural language user input;

processing, using a machine learning (ML) component, the first embedding data and the second embedding data to determine the second natural language user input is associated with the first content data; and

based on determining that the second natural language user input is associated with the first content data, performing a second action responsive to the second natural language user input.

6 . The computer-implemented method of claim 5 , further comprising:

processing, using an encoder, the second representation of the second natural language user input to generate the second embedding data representing the at least one feature of the second natural language user input;

processing, using the ML component, the first embedding data and the second embedding data to determine a similarity between the first embedding data and the second embedding data,

determining the similarity represents a likelihood that the second embedding data is requesting performance of the second action with respect to the first content data; and

determining, based on the similarity, that the second natural language user input relates to the first content data.

7 . The computer-implemented method of claim 5 , further comprising:

determining an object represented in the first content data;

determining second content data including a third representation of the object; and

based on the second content data including the third representation of the object, determining that the second content data is to be displayed in response to the first natural language user input.

8 . The computer-implemented method of claim 5 , further comprising:

determining second content data that was displayed prior to receiving the first representation of the first natural language user input, wherein the first content data and the second content data are determined based on first data associating the first content data and the second content data with a first device, and the first natural language user input being received from the first device;

determining third embedding data associated with the second content data, the third embedding data representing at least one feature of the second content data;

processing, using the machine learning (ML) component, the first representation of the first natural language user input and the first embedding data to determine a first similarity;

processing, using the ML component, the first representation of the first natural language user input and the third embedding data to determine a second similarity; and

based on the first similarity and the second similarity, determining that the first natural language user input relates to the first content data instead of the second content data.

9 . The computer-implemented method of claim 8 , further comprising:

based on determining that the first natural language user input refers to the first content data instead of the second content data, generating natural language understanding (NLU) output data including at least a first NLU hypothesis associated with the first embedding data and the third embedding data.

10 . The computer-implemented method of claim 5 , further comprising:

determining the first embedding data represents at least one of a color or a position of an object represented in the first content data.

11 . The computer-implemented method of claim 5 , further comprising:

determining a first automatic speech recognition (ASR) hypothesis including a first transcript of the first natural language user input; and

using a machine learning (ML) component, processing the first ASR hypothesis and the first embedding data to determine a second ASR hypothesis including a second transcript of the first natural language user input.

12 . The computer-implemented method of claim 5 , wherein:

the first content data is video data, and

the first embedding data corresponds to a first frame of the video data.

13 . A computing system comprising:

at least one processor; and

at least one memory comprising instructions that, when executed by the at least one processor, cause the computing system to:

receive a first representation of a first natural language user input;

determine first content data that was displayed prior to receiving the first representation of the first natural language user input;

determine first embedding data associated with the first content data, the first embedding data representing at least one feature of the first content data;

use the first embedding data, determining that the first natural language user input refers to the first content data;

based on determining that the first natural language user input refers to the first content data, perform a first action responsive to the first natural language user input;

receive a second representation of a second natural language user input;

determine second embedding data representing at least one feature of the second natural language user input;

process, using a machine learning (ML) component, the first embedding data and the second embedding data to determine the second natural language user input is associated with the first content data; and

based on determination that the second natural language user input is associated with the first content data, perform a second action responsive to the second natural language user input.

14 . The computing system of claim 13 , further comprising:

process, using an encoder, the second representation of the second natural language user input to generate the second embedding data representing at least one feature of the second natural language user input;

process, using the ML component, the first embedding data and the second embedding data to determine a similarity between the first embedding data and the second embedding data,

determine the similarity represents a likelihood that the second embedding data is requesting performance of the second action with respect to the first content data; and

determine, based on the similarity, that the second natural language user input relates to the first content data.

15 . The computing system of claim 13 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine an object represented in the first content data;

determine second content data including a third representation of the object; and

based on the second content data including the third representation of the object, determine that the second content data is to be displayed in response to the first natural language user input.

16 . The computing system of claim 13 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine second content data that was displayed prior to receiving the first representation of the first natural language user input, wherein the first content data and the second content data are determined based on first data associating the first content data and the second content data with a first device, and the first natural language user input being received from the first device;

determine third embedding data associated with the second content data, the third embedding data representing at least one feature of the second content data;

process, using the ML component, the first representation of the first natural language user input and the first embedding data to determine a first similarity;

process, using the ML component, the first representation of the first natural language user input and the third embedding data to determine a second similarity; and

based on the first similarity and the second similarity, determine that the first natural language user input relates to the first content data instead of the second content data.

17 . The computing system of claim 16 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

based on determining that the first natural language user input refers to the first content data instead of the second content data, generate natural language understanding (NLU) output data including at least a first NLU hypothesis associated with the first embedding data and the third embedding data.

18 . The computing system of claim 13 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine the first embedding data represents at least one of a color or a position of an object represented in the first content data.

19 . The computing system of claim 13 , wherein the at least one memory further comprises instructions that, when executed by the at least one processor, further cause the computing system to:

determine a first automatic speech recognition (ASR) hypothesis including a first transcript of the first natural language user input; and

process, using a machine learning (ML) component, the first ASR hypothesis and the first embedding data to determine a second ASR hypothesis including a second transcript of the first natural language user input.

20 . The computing system of claim 13 , wherein:

the first content data is video data, and

the first embedding data corresponds to a first frame of the video data.

Assignments (1)

Continuity (1)

References Cited (44)