Granted Patent B2

US 12,197,862 · App. 17/980,423 · Granted Jan 14, 2025

Inventor: Seongmin Park (Seoul, KR)

Assignee: ActionPower Corp.

G06F40/279G06F16/3344G06F40/166G06F40/53G10L15/187G10L15/22

Abstract

Claims (69)

1. A method for identifying a word corresponding to a target word, the method is performed by a computing device including at least one processor, the method comprising:

obtaining original voice data of a user via an input module, wherein a transcription result is generated from the original voice data;

receiving a set of modified text associated with a set of portions of the transcription result generated from the original voice data;

training a speech recognition model for the user using the set of modified text paired with a set of portions of the original voice data respectively corresponding to the set of portions of the transcription result, wherein the speech recognition model is trained using a neural network comprising a plurality of layers through which the set of modified text paired with the set of portions of the original voice data is processed;

outputting, by the speech recognition model that was trained, text information comprising a set of words;

determining the target word from the set of words included in the text information, including at least one of:

determining the target word based on a keyword dictionary of a predetermined field,

determining a word of which frequency is equal to or more than a threshold in the text information as the target word, or

determining the target word based on input information received through a user interface;

determining a threshold for an edit distance associated with the target word by multiplying a length of the target word by a predetermined ratio threshold;

determining, from the set of words included in the text information, at least one word of which the edit distance from the target word is equal to or less than the threshold; and

identifying, from the at least one word, the word corresponding to the target word.

2. The method of claim 1 , wherein the text information includes text information generated through speech recognition, and

the determining of the target word includes determining the target word based on a confidence related to the speech recognition.

3. The method of claim 1 , wherein the text information includes text information generated through speech recognition, and

the identifying of the word corresponding to the target word further includes:

calculating a similarity between voice information associated with the target word and voice information associated with the text information, and

identifying the word corresponding to the target word by additionally considering the calculated similarity.

4. The method of claim 1 , wherein the determining of the threshold for the edit distance further includes determining an absolute threshold, and

the determining of the at least one word of which the edit distance from the target word is equal to or less than the threshold further includes determining a word of which the edit distance from the target word is equal to or less than the threshold, and equal to or less than the absolute threshold.

5. The method of claim 1 , wherein the determining of the at least one word of which the edit distance from the target word is equal to or less than the threshold includes:

dividing the target word into k+1 elements when the determined threshold is k,

searching the k+1 elements in the text information,

identifying the at least one word based on the searched element when at least one element among the k+1 elements is searched in the text information, and

determining that there is no word of which the edit distance is equal to or less than the threshold when the k+1 elements are not searched in the text information.

6. The method of claim 1 , wherein the word corresponding to the target word is identified based on a pronunciation or a text, and

the edit distance is calculated based on the pronunciation or the text.

7. The method of claim 6 , wherein the edit distance is determined based on the pronunciation, and the determining of the at least one word of which the edit distance from the target word is equal to or less than the threshold includes:

converting words included in the text information and the target word into pronunciation symbols,

identifying pronunciation symbols of which the edit distance from the pronunciation symbols for the target word is equal to or less than the threshold among the pronunciation symbols for the text information, and

determining the at least one word based on the identified pronunciation symbols.

8. The method of claim 7 , wherein the converting of the words included in the text information and the target word into the pronunciation symbols includes:

converting the words included in the text information and the target word into an International Phonetic Alphabet (IPA) alphabet,

converting the converted IPA alphabets into Roman notations,

separating the converted Roman notations into grapheme units, and

converting the separated grapheme unit into a phoneme unit.

9. The method of claim 6 , wherein the edit distance is determined based on the text, and the determining of the at least one word of which the edit distance from the target word is equal to or less than the threshold includes:

converting the words included in the text information and the target word into the grapheme units,

identifying graphemes of which the edit distance from the graphemes for the target word is equal to or less than the threshold among the graphemes for the text information, and

determining the at least one word based on the identified graphemes.

10. The method of claim 6 , wherein the determining of the at least one word of which the edit distance from the target word is equal to or less than the threshold includes converting the words included in the text information and the target word into the pronunciation symbols or the graphemes, and

the converting of the words included in the text information and the target word into the pronunciation symbols or the graphemes includes selecting whether to consider at least one of a spacing or the sentence symbol.

11. A computer program stored in a non-transitory computer-readable storage medium, wherein the computer program executes the following operations for identifying a word corresponding to a target word when the computer program is executed by one or more processors, the operations comprising:

an operation of obtaining original voice data of a user via an input module, wherein a transcription result is generated from the original voice data;

an operation of receiving a set of modified text associated with a set of portions of the transcription result generated from the original voice data;

an operation of training a speech recognition model for the user using the set of modified text paired with a set of portions of the original voice data respectively corresponding to the set of portions of the transcription result, wherein the speech recognition model is trained using a neural network comprising a plurality of layers through which the set of modified text paired with the set of portions of the original voice data is processed;

an operation of outputting, by the speech recognition model that was trained, text information comprising a set of words;

an operation of determining the target word from the set of words included in the text information, including at least one of:

an operation of determining the target word based on a keyword dictionary of a predetermined field,

an operation of determining a word of which frequency is equal to or more than a threshold in the text information as the target word, or

an operation of determining the target word based on input information received through a user interface;

an operation of determining a threshold for an edit distance associated with the target word by multiplying a length of the target word by a predetermined ratio threshold;

an operation of determining, from the set of words included in the text information, at least one word of which the edit distance from the target word is equal to or less than the threshold; and

an operation of identifying, from the at least one word, the word corresponding to the target word.

12. A computing device for identifying a word corresponding to a target word in text information, comprising:

at least one processor; and

a memory,

wherein at least one processor is configured to:

obtain original voice data of a user via an input module, wherein a transcription result is generated from the original voice data,

receive a set of modified text associated with a set of portions of the transcription result generated from the original voice data,

train a speech recognition model using modified text data paired with a set of portions of the original voice data respectively corresponding to the set of portions of the transcription result, wherein the speech recognition model is trained using a neural network comprising a plurality of layers through which the set of modified text paired with the set of portions of the original voice data is processed,

output, by the speech recognition model that was trained, text information comprising a set of words,

determine the target word from the set of words included in the text information, including at least one of:

determine the target word based on a keyword dictionary of a predetermined field,

determine a word of which frequency is equal to or more than a threshold in the text information as the target word, or

determine the target word based on input information received through a user interface,

determine a threshold for an edit distance associated with the target word by multiplying a length of the target word by a predetermined ratio threshold,

determine, from the set of words included in the text information, at least one word of which the edit distance from the target word among words included in text information is equal to or less than the threshold, and

identify, from the at least one word, the word corresponding to the target word.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (83)