# US 2021/0065704 A1 · Method and apparatus for providing speech recognition service

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0065704 A1
- Document ID: US20210065704A1
- Status: Published application
- Kind: A1
- Application number: 16/739,811
- Filed: 2020-01-10
- Published: 2021-03-04
- Earliest claimed: 2019-08-29
- Assignees: LG Electronics Inc.
- Applicants: LG Electronics Inc.
- Inventors: Da Hae KIM
- Web page: https://patentgrep.com/patent/US20210065704A1

## CPC

- G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue
- G06F 3/167: {Audio in a user interface, e.g. using voice commands for navigating, audio feedback}
- G06F 40/295: Named entity recognition
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G10L 15/16: using artificial neural networks
- G10L 15/1822: {Parsing for meaning understanding}
- G10L 15/30: Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
- G06N 3/08: Learning methods
- G10L 2015/223: {Execution procedure of a spoken command}
- G10L 2015/226: {using non-speech characteristics}
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Disclosed are a method for providing a speech recognition service and a speech recognition apparatus, which may perform speech recognition by executing an artificial intelligence (AI) algorithm and/or a machine learning algorithm, which are mounted therein, so that a speech recognition apparatus and a server may communicate with each other in a 5G communication environment. The method and the speech recognition apparatus provide a response based on a user's intention analysis with respect to the ambiguous utterance of the user.

## Claims

### Claim 1

A method for providing a speech recognition service, comprising:
receiving a speech input of a user;
obtaining a plurality of candidate actions extracted from the speech input;
deciding relevance between the speech input and each candidate action of the plurality of candidate actions based on current context information of the user; and
deciding a final action of the plurality of candidate actions based on the relevance,
wherein the deciding the relevance comprises:
deciding a weight of each candidate action for each type of each context information by analyzing accumulated context information with respect to the user; and
calculating the relevance by combining the weights for each candidate action.

### Claim 2 (depends on 1)

The method of claim 1,
wherein the obtaining the plurality of candidate actions comprises:
transmitting the speech input to an external server; and
receiving the plurality of candidate actions extracted by performing a natural language processing for a text representing the speech input by the external server.

### Claim 3 (depends on 1)

The method of claim 1,
wherein the each candidate action of the plurality of candidate actions comprises a named entity that is a target of the action and domain information having classified a category of the action.

### Claim 4 (depends on 3)

The method of claim 3,
wherein the each candidate action of the plurality of candidate actions are distinguished from each other by a combination of the named entity and the domain information.

### Claim 5 (depends on 1)

The method of claim 1,
wherein the current context information comprises at least one of information on a time of the speech input uttered by the user, a place of the speech input uttered by the user, whether the user is moving, a moving speed of the user, or a device being used by the user.

### Claim 6 (depends on 1)

The method of claim 1,
wherein the accumulated context information with respect to the user comprises at least one of information on a time received by a command, a place received by the command, a frequency performed by an action, whether the user was moving when the command was received, a moving speed of the user when the command is received, or a device used in receiving the command, for each action performed by the user's command.

### Claim 7 (depends on 1)

The method of claim 1,
wherein the deciding the weight comprises:
deciding the frequency of having performed each candidate action by analyzing the accumulated context information with respect to the user; and
deciding a weight of each candidate action with respect to the type of each context information based on the frequency.

### Claim 8 (depends on 7)

The method of claim 7,
wherein a first type of the context information is time information of the speech input uttered by the user,
wherein the deciding the frequency comprises:
deciding the frequency by counting the number of times of each candidate action performed by the user for each predetermined time unit, and
wherein the deciding the weight of each candidate action with respect to the type of each context information based on the frequency comprises:
normalizing the number of times that has performed the each candidate action with respect to the entire time period, and deciding a result value of the normalization corresponding to the time information of having uttered the speech input from the current context information as a weight for the first type of the each candidate action.

### Claim 9 (depends on 7)

The method of claim 7,
wherein a second type of the context information is place information of the speech input uttered by the user,
wherein the deciding the frequency comprises:
deciding the frequency by counting the number of times of each candidate action performed by the user in at least one place where the user stays for a predetermined time or more, and
wherein the deciding the weight of each candidate action with respect to the type of each context information based on the frequency comprises:
normalizing the number of times that has performed the each candidate action with respect to all places recorded in the accumulated context information, and deciding a result value of the normalization corresponding to the place of having uttered the speech input from the current context information as a weight for the second type of the each candidate action.

### Claim 10 (depends on 7)

The method of claim 7,
wherein a third type of the context information is frequency information of the action performed by the user,
wherein the deciding the frequency comprises:
deciding the frequency for each action performed by the user by counting the number of times performed for each action performed by the user for a predetermined time period, and
wherein the deciding the weight of each candidate action with respect to the type of each context information based on the frequency comprises:
normalizing the frequency for each action, and deciding a result value of the normalization corresponding to the each candidate action as a weight for the third type of the each candidate action.

### Claim 11 (depends on 1)

The method of claim 1,
wherein the deciding the relevance is repeatedly performed for each predetermined time period.

### Claim 12 (depends on 1)

The method of claim 1,
wherein the deciding the final action comprises deciding a candidate action having the maximum relevance as the final action.

### Claim 13

A speech recognition apparatus, comprising:
a microphone configured to receive a speech input of a user; and
a processor configured to decide one of a plurality of candidate actions extracted from the speech input as a final action,
wherein the processor is configured to perform:
an operation of deciding relevance between the speech input and each candidate action of the plurality of candidate actions based on current context information of the user; and
an operation of deciding the final action of the plurality of candidate actions based on the relevance, and
wherein the operation of deciding the relevance comprises:
an operation of deciding a weight of the each candidate action for each type of each context information by analyzing accumulated context information with respect to the user; and
an operation of calculating the relevance by combining the weights for the each candidate action.

### Claim 14 (depends on 13)

The speech recognition apparatus of claim 13,
wherein the each candidate action of the plurality of candidate actions are distinguished from each other by a combination of a named entity that is a target of the action and domain information having classified a category of the action.

### Claim 15 (depends on 13)

The speech recognition apparatus of claim 13,
wherein the operation of deciding the weight comprises:
an operation of deciding a frequency of having performed the candidate action by analyzing the accumulated context information; and
an operation of deciding the weight of the each candidate action with respect to the type of each context information based on the frequency.

### Claim 16 (depends on 13)

The speech recognition apparatus of claim 13,
wherein the operation of calculating the relevance comprises an operation of summing the weights for the each candidate action.
