Artificial intelligence apparatus for performing speech recognition and method thereof

Patent metadata

Field	Value
Publication number	US-11289074-B2
Application number	US-201916653569-A
Country	US
Kind code	B2
Filing date	Oct 15, 2019
Priority date	Oct 1, 2019
Publication date	Mar 29, 2022
Grant date	Mar 29, 2022

Abstract

Disclosed herein is an artificial intelligence apparatus for performing speech recognition including a microphone configured to receive a speech command of a user, a learning processor configured to determine anaphora included in text data corresponding to the speech command using an anaphora recognition model for determining anaphora included in predetermined text data, and a processor configured to specify an object referred to by the determined anaphora based on context information including information input to or output from the artificial intelligence apparatus, determine a response to the speech command based on the specified object, and control the artificial intelligence apparatus according to the determined response.

First claim

What is claimed is:

1. An artificial intelligence apparatus for performing speech recognition, the artificial intelligence apparatus comprising:
a microphone configured to receive a speech command of a user;
a processor configured to:
determine anaphora included in text data corresponding to the speech command using an anaphora recognition model for determining anaphora included in predetermined text data;
specify an object referred to by the determined anaphora based on context information including information input to or output from the artificial intelligence apparatus, screen context information including at least information on output content output via a display of the artificial intelligence apparatus and information on selected content selected by the user, and conversation context information which is information on a conversation history between the user and the artificial intelligence apparatus via a speech command and a response to the speech command,
wherein the anaphora includes at least one of adjacent anaphora referring to an object adjacent to a person or a thing or non-adjacent anaphora referring to an object which is not adjacent to a person or a thing, wherein the adjacent anaphora or the non-adjacent anaphora included in the text data corresponding to the speech command is determined using the anaphora recognition model,
determine a response to the speech command based on the specified object, and
control the artificial intelligence apparatus according to the determined response.

2. The artificial intelligence apparatus of claim 1 , wherein the processor is further configured to specify the object referred to by the anaphora based on the selected content when at least one selected content is present.

3. The artificial intelligence apparatus of claim 1 , wherein, if the output content is full-screen content output on a full screen of the display, the processor is further configured to specify the object referred to by the anaphora based on the full-screen content.

4. The artificial intelligence apparatus of claim 1 , wherein the processor is further configured to specify the object referred to by the determined adjacent anaphora based on the screen context information.

5. The artificial intelligence apparatus of claim 1 , wherein the processor is further configured to specify the object referred to by the determined non-adjacent anaphora based on the conversation context information.

6. The artificial intelligence apparatus of claim 5 ,
wherein the conversation context information includes information on a candidate of an object referred to by anaphora included in text data corresponding to each speech command included in the conversation history, and
wherein the processor is further configured to specify the object referred to by the determined non-adjacent anaphora based on the candidate of the object referred to by the anaphora.

7. The artificial intelligence apparatus of claim 1 , wherein the anaphora recognition model is a neural network trained by a machine learning algorithm or a deep learning algorithm to output an anaphora tagging value of each character or word included in the predetermined text data when the predetermined text data is input to an input layer.

8. A speech recognition method performed by an artificial intelligence apparatus, the speech recognition method comprising:
receiving a speech command of a user;
determining anaphora included in text data corresponding to the speech command using an anaphora recognition model for determining anaphora included in predetermined text data;
specifying an object referred to by the determined anaphora based on context information including information input to or output from the artificial intelligence apparatus, screen context information including at least information on output content output via a display of the artificial intelligence apparatus and information on selected content selected by a user, and conversation context information which is information on a conversation history between the user and the artificial intelligence apparatus via a speech command and a response to the speech command, wherein the anaphora includes at least one of adjacent anaphora referring to an object adjacent to a person or a thing or non-adjacent anaphora referring to an object which is not adjacent to a person or a thing, wherein the adjacent anaphora or the non-adjacent anaphora included in the text data corresponding to the speech command is determined using the anaphora recognition model;
determining a response to the speech command based on the specified object; and
controlling the artificial intelligence apparatus according to the determined response.

9. The speech recognition method of claim 8 , wherein the specifying of the object includes specifying the object referred to by the anaphora based on the selected content when at least one selected content is present.

10. The speech recognition method of claim 8 , wherein the specifying of the object includes, if the output content is full-screen content output on a full screen of the display, specifying the object referred to by the anaphora based on the full-screen content.

11. The speech recognition method of claim 8 , wherein the specifying of the object includes specifying the object referred to by the determined adjacent anaphora based on the screen context information.

12. The speech recognition method of claim 8 , wherein the specifying of the object includes specifying the object referred to by the determined non-adjacent anaphora based on the conversation context information.

13. The speech recognition method of claim 12 ,
wherein the conversation context information includes information on a candidate of an object referred to by anaphora included in text data corresponding to each speech command included in the conversation history, and
wherein the specifying of the object includes specifying the object referred to by the determined non-adjacent anaphora based on the candidate of the object referred to by the anaphora.

14. The speech recognition method of claim 8 , wherein the anaphora recognition model is a neural network trained by a machine learning algorithm or a deep learning algorithm to output an anaphora tagging value of each character or word included in the predetermined text data when the predetermined text data is input to an input layer.

Assignees

• Lg Electronics Inc

Inventors

• Lee Kwangyong

Classifications

• G06N3/09

  Supervised learning · CPC title

• G10L2015/223

  Execution procedure of a spoken command · CPC title

• G10L15/1822 Primary

  Parsing for meaning understanding · CPC title

• G10L15/16 Primary

  using artificial neural networks · CPC title

• G10L15/22 Primary

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title