Language model error correction method for improving voice recognition capability

Abstract

The invention discloses a language model error correction method, device and equipment for improving voice recognition capability and a storage medium. The method comprises the steps: obtaining text information obtained through voice recognition as a to-be-corrected text, obtaining each word in the to-be-corrected text and the score of each word, and determining the word with the score lower thana corresponding threshold as a to-be-corrected word; determining a word before the to-be-corrected word in the to-be-corrected text as a to-be-corrected precursor word, and determining an alternativeword matched with the to-be-corrected precursor word in a word bank as a to-be-corrected alternative word; and respectively replacing to-be-corrected words in the to-be-corrected text with the to-be-corrected alternative words, obtaining scores of to-be-corrected alternative words, and determining a to-be-corrected alternative word with the highest score as a word obtained by correcting the to-be-corrected word, wherein the score of any word is the probability that any word appears on the premise that the word located in front of the any word in the to-be-corrected text appears, and the word bank comprises a plurality of precursor words and a plurality of alternative words matched with the precursor word. The voice recognition accuracy can be improved.

Description

technical field [0001] The present invention relates to the technical field of speech recognition, and more specifically, relates to a language model error correction method, device, device and storage medium for improving speech recognition ability. Background technique [0002] In order to improve customer experience, many industries currently use intelligent equipment to respond to voices from customers and realize corresponding operations; when implementing voice recognition, voice recognition models are usually used to perform corresponding voice recognition, but the inventors found that the existing technology In the solution, after the speech is recognized to obtain the corresponding text information, the text information obtained by the speech recognition may be inconsistent with the text information to be expressed by the speech, resulting in low accuracy of speech recognition. Contents of the invention [0003] The purpose of the present invention is to provide a...

AI Technical Summary

Problems solved by technology

[0002] In order to improve customer experience, many industries currently use intelligent equipment to respond to voices from customers and realize corresponding operations; when implementing voice recognition, voice recognition models are usually used to perform corresponding voice recognition, but the inventors found that the existing technology After the solution recognizes the speech to obtain the corresponding text information, the text information obtained by the speech recognition may be inconsistent with the text information to be expressed by the speech, resulting in low accuracy of speech recognition

Images