Text processing method and apparatus, electronic device, and medium

Patent metadata

Field	Value
Publication number	US-2023326466-A1
Application number	US-202118043514-A
Country	US
Kind code	A1
Filing date	Aug 24, 2021
Priority date	Aug 31, 2020
Publication date	Oct 12, 2023
Grant date	—

Abstract

Provided are a text processing method and apparatus, an electronic device, and a medium. The method includes the following: target text information generated based on audio information is acquired; a to-be-error-corrected word in the target text information and a target candidate replacement word corresponding to the to-be-error-corrected word are determined; and a target replacement word corresponding to the to-be-error-corrected word is determined according to the target candidate replacement word, and the target text information is updated based on the target replacement word.

First claim

1 . A text processing method, comprising:
acquiring target text information generated based on audio information; determining a to-be-error-corrected word in the target text information and a target candidate replacement word corresponding to the to-be-error-corrected word; and determining, according to the target candidate replacement word, a target replacement word corresponding to the to-be-error-corrected word, and updating the target text information based on the target replacement word.

2 . The method according to claim 1 , before acquiring the target text information generated based on the audio information, further comprising:
collecting the audio information of a speaker and converting the audio information to corresponding text information; and generating, according to the text information, a speech timestamp corresponding to the speaker, and an identifier of the speaker, current text content displayed on a client, and determining the target text information based on the current text content.

3 . The method according to claim 2 , wherein acquiring the target text information generated based on the audio information comprises:
determining a timestamp of text content without error correction among all text content, and acquiring text content without error correction within a preset duration based on the timestamp; and determining the target text information based on the text content without error correction within the preset duration.

4 . The method according to claim 3 , wherein the all text content is determined based on text information displayed in a preset region of the client, or the all text content is retrieved from a speech-to-text module.

5 . The method according to claim 1 , wherein determining the to-be-error-corrected word in the target text information and the target candidate replacement word corresponding to the to-be-error-corrected word comprises:
determining the to-be-error-corrected word in the target text information and the target candidate replacement word corresponding to the to-be-error-corrected word in a correction manner corresponding to a correction type, to determine the target replacement word based on the target candidate replacement word, wherein the error correction type comprises a type of text-pronunciation-based error correction and a type of text-content-based error correction.

6 . The method according to claim 5 , wherein the error correction type comprises the type of text-pronunciation-based error correction, and adopting the error correction manner corresponding to the error correction type to determine the to-be-error-corrected word in the target text information and the target candidate replacement word corresponding to the to-be-error-corrected word comprises:
acquiring a pronunciation of each piece of text in the target text information; determining, according to the pronunciation of the each piece of text and hot words pre-stored in a hot word dictionary, whether a target hot word corresponding to the pronunciation of the each piece of text exists in the target text information, wherein the hot word dictionary is configured to store a plurality of hot words, and the plurality of hot words are determined based on audio information and text information that are collected in a real-time interactive process; and in response to the target hot word corresponding to the pronunciation of the each piece of text existing in the target text information, determining the to-be-error-corrected word in the target text information according to the target hot word, and determining the target candidate replacement word based on the to-be-error-corrected word.

7 . The method according to claim 6 , wherein the type of text-pronunciation-based error correction comprises a type of pinyin-based error correction.

8 . The method according to claim 6 , wherein determining, according to the target candidate replacement word, the target replacement word corresponding to the to-be-error-corrected word, and updating the target text information based on the target replacement word comprises:
acquiring a first to-be-processed sentence to which the to-be-error-corrected word belongs in the target text information, and updating the first to-be-processed sentence based on the target candidate replacement word to acquire a second to-be-processed sentence; determining a perplexity value of the second to-be-processed sentence; in response to the perplexity value being greater than or equal to a preset perplexity threshold, determining the target replacement word according to the to-be-error-corrected word; and in response to the perplexity value being less than the preset perplexity threshold, determining the target replacement word according to the target candidate replacement word.

9 . The method according to claim 5 , wherein the error correction type comprises the type of text-content-based error correction, and adopting the error correction manner corresponding to the error correction type to determine the to-be-error-corrected word in the target text information and the target candidate replacement word corresponding to the to-be-error-corrected word so as to determine the target replacement word based on the target candidate replacement word comprises at least one of:
performing matching for text content in the target text information based on a pre-determined confusion word lexicon to determine a confusion word in the target text information, and determining the to-be-error-corrected word and the corresponding target candidate replacement word based on the confusion word; or determining a suspected to-be-error-corrected word in the target text information based on a pre-determined common word lexicon, and determining the to-be-error-corrected word and the corresponding target candidate replacement word based on the suspected to-be-error-corrected word.

10 . The method according to claim 9 , wherein performing the matching for the text content in the target text information based on the pre-determined confusion word lexicon to determine the confusion word in the target text information, and determining the to-be-error-corrected word and the corresponding target candidate replacement word based on the confusion word comprises:
determining the confusion word in the target text information based on the pre-determined confusion word lexicon, and taking a word in the pre-determined confusion word lexicon corresponding to the confusion word as the target candidate replacement word; and determining, according to the target candidate replacement word, the target replacement word corresponding to the to-be-error-corrected word, and updating the target text information based on the target replacement word comprises: determining the target replacement word according to the target candidate replacement word; and updating, based on the target replacement word, the to-be-error-corrected word in the target text information corresponding to the target replacement word.

11 . The method according to claim 9 , before determining the suspected to-be-error-corrected word in the target text information based on the pre-determined common word lexicon, and determining the to-be-error-corrected word and the corresponding target candidate replacement word based on the suspected to-be-error-corrected word, further comprising:
segmenting a sentence in the target text information to acquire at least one key word to determine the to-be-error-corrected word from the at least one key word.

12 . The method according to claim 11 , wherein determining the to-be-error-corrected word in the target text information and the target candidate replacement word cor

Assignees

• Beijing Bytedance Network Tech Co Ltd

Inventors

• Yang Jingsheng
• Du Chunsai
• Xu Wenming
• Zhao Li
• Han Xiao

Classifications

• G10L15/22

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

• G10L17/20 Primary

  Pattern transformations or operations aimed at increasing system robustness, e.g. against channel noise or different working conditions · CPC title

• G10L17/02

  Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction · CPC title

• G10L17/14

  Use of phonemic categorisation or speech recognition prior to speaker recognition or verification · CPC title

• G06F40/232 Primary

  Orthographic correction, e.g. spell checking or vowelisation · CPC title