Granted Patent B2

US 12,597,413 · App. 18/105,011 · Granted Apr 7, 2026

Inventors: Soyoon Park (Suwon-si, KR); Sungjun Lim (Suwon-si, KR); Jonghyun Kim (Suwon-si, KR); Jiwan Kim (Suwon-si, KR); Hakjung Kim (Suwon-si, KR); Hyunkyung Kim (Suwon-si, KR); Sohyun Park (Suwon-si, KR); Indong Lee (Suwon-si, KR)

Assignee: SAMSUNG ELECTRONICS CO., LTD.

G10L13/086G10L15/005G10L15/02G10L15/26

Abstract

Claims (79)

1 . An electronic apparatus comprising:

a microphone;

a communication interface comprising communication circuitry;

a memory configured to store a first encoder corresponding to a first language and a first decoder corresponding to the first language; and

a processor configured to:

based on a user voice in the first language being received through the microphone, acquire a first text in the first language corresponding to the user voice,

acquire a first feature vector by inputting the first text in the first language to the first encoder,

control the communication interface to transmit the first feature vector to an external device, and

based on a second feature vector being received from the external device through the communication interface, acquire text in the first language corresponding to the second feature vector by inputting the second feature vector to the first decoder,

wherein the external device comprises a second encoder corresponding to a second language, and

wherein the second encoder is configured to output the second feature vector based on a second text corresponding to a user voice in the second language being input.

2 . The electronic apparatus of claim 1 , wherein the first encoder comprises a model learned based on the first text in the first language as input data and a feature vector on a vector space as output data, and

wherein the second encoder corresponding comprises a model learned based on the second text in the second language having a similarity greater than or equal to a threshold value with the first text as input data and a feature vector in a vector space as output data.

3 . The electronic apparatus of claim 1 , wherein the first decoder comprises a model learned based on a feature vector on a vector space as input data and the first text in the first language as output data, and

wherein a second decoder corresponding to the second language different from the first language comprises a model learned based on a feature vector in a vector space as input data and the second text in the second language having a similarity greater than or equal to a threshold value with the first text as output data.

4 . The electronic apparatus of claim 3 , wherein the external device includes the second decoder corresponding to the second language, and

wherein the second decoder is configured to output text in the second language based on the first feature vector received from the electronic apparatus being input.

5 . The electronic apparatus of claim 3 , wherein the processor is further configured to:

identify whether the external device includes the second decoder corresponding to the second language by communicating with the external device via the communication interface,

based on identifying that the external device includes the second decoder, control the communication interface to transmit the first feature vector to the external device, and

based on identifying that the external device does not include the second decoder, control the communication interface to transmit the first feature vector to a server.

6 . The electronic apparatus of claim 1 , further comprising:

a speaker,

wherein the processor is further configured to:

acquire a sound of the first language corresponding to the text in the first language using text to speech, and

output the sound of the first language through the speaker.

7 . The electronic apparatus of claim 1 , wherein the memory further comprises a compressor and a decompressor,

wherein the processor is further configured to:

compress the first feature vector based on the compressor,

control the communication interface to transmit the compressed first feature vector to the external device,

decompress a compressed second feature vector based on the decompressor based on the compressed second feature vector being received from the external device, and

input the decompressed second feature vector to the first decoder.

8 . The electronic apparatus of claim 1 , wherein the first encoder and the first decoder are included in a neural machine translation model,

wherein the neural machine translation model is configured to:

based on the user voice being input, acquire the first feature vector by converting the text corresponding to the user voice into a vector value, and

based on the second feature vector being input, convert the second feature vector into text in the first language.

9 . A method for controlling an electronic apparatus, the method comprising:

based on a user voice in a first language being received, acquiring text in the first language corresponding to the user voice;

acquiring a first feature vector by inputting the text in the first language to a first encoder;

transmitting the first feature vector to an external device; and

based on a second feature vector being received from the external device, acquiring text in the first language corresponding to the second feature vector by inputting the second feature vector to a first decoder corresponding to the first language,

wherein the external device comprises a second encoder corresponding to a second language, and

wherein the second encoder is configured output the second feature vector based on a second text corresponding to the user voice in the second language being input to the second encoder.

10 . The method of claim 9 , wherein the first encoder comprises a model learned based on the first text in the first language as input data and a feature vector on a vector space as output data, and

wherein the second encoder comprises a model learned based on the second text in the second language having a similarity greater than or equal to a threshold value with the first text as input data and a feature vector in a vector space as output data.

11 . The method of claim 10 , wherein the first decoder comprises a model learned based on a feature vector on a vector space as input data and the first text in the first language as output data, and

wherein a second decoder corresponding to the second language different from the first language comprises a model learned based on a feature vector in a vector space as input data and the second text in the second language having a similarity greater than or equal to a threshold value with the first text as output data.

12 . The method of claim 11 , wherein the external device includes the second decoder corresponding to the second language, and

wherein the second decoder is configured to output text in the second language based on the first feature vector received from the electronic apparatus being input.

13 . The method of claim 11 , further comprising:

identifying whether the external device includes the second decoder corresponding to the second language by communicating with the external device;

based on identifying that the external device includes the second decoder, transmitting the first feature vector to the external device; and

based on identifying that the external device does not include the second decoder, transmitting the first feature vector to a server.

14 . The method of claim 9 , further comprising:

acquiring a sound of the first language corresponding to the text in the first language using text to speech; and

outputting the sound of the first language through a speaker.

15 . The method of claim 9 , further comprising:

compressing the first feature vector based on a compressor;

transmitting the compressed first feature vector to the external device through a communication interface;

decompressing a compressed second feature vector based on a decompressor based on the compressed second feature vector being received from the external device; and

inputting the decompressed second feature vector to the first decoder.

16 . The method of claim 9 , wherein the first encoder and the first decoder are included in a neural machine translation model,

wherein the neural machine translation model is configured to:

based on the user voice being input, acquire the first feature vector by converting the text corresponding to the user voice into a vector value; and

based on the second feature vector being input, convert the second feature vector into text in the first language.

17 . An electronic apparatus comprising:

a microphone;

a communication interface comprising communication circuitry;

a memory configured to store a first encoder and a first decoder corresponding to a first language; and

a processor configured to:

based on a user voice in the first language being received through the microphone, acquire a first text in the first language corresponding to the user voice;

acquire a first feature vector by inputting the first text in the first language to the first encoder;

control the communication interface to transmit the first feature vector to an external device;

based on a second feature vector being received from the external device through the communication interface, acquire text in the first language corresponding to the second feature vector by inputting the second feature vector to the first decoder;

identify whether the external device comprises a second decoder corresponding to a second language different from the first language by communicating with the external device via the communication interface;

based on identifying that the external device comprises the second decoder, transmit the first feature vector to the external device; and

based on identifying that the external device does not comprise the second decoder, transmit the first feature vector to a server.

18 . The electronic apparatus of claim 17 , wherein the first decoder comprises a model learned based on a feature vector on a vector space as input data and the first text in the first language as output data, and

wherein the second decoder comprises a model learned based on a feature vector in a vector space as input data and a second text in the second language having a similarity greater than or equal to a threshold value with the first text as output data.

Assignments (1)

Priority Claims (2)

Continuity (2)

References Cited (175)