Statement error correction method and device after speech recognition, equipment and storage medium

Abstract

The embodiment of the invention discloses a statement error correction method and device after speech recognition, equipment and a storage medium. According to the technical scheme provided by the embodiment of the invention, the method comprises the steps of: recognizing the first occurrence probability of each character in the to-be-corrected text through the language model, determining the recognized error word in the to-be-corrected text according to the first occurrence probability, determining the model candidate word by utilizing the language model, determining the homophone candidate word according to the pinyin and tone of the recognized error word, further determining a first sequence and a second sequence between a model candidate word and a homophone candidate word, determining a candidate sequence between the model candidate word and the homophone candidate word according to the first sequence and the second sequence, determining an error correction candidate word according to the candidate sequence,replacing a recognition error word in a to-be-corrected text with the error correction candidate word, and directly docking and modifying the voice recognition result in a non-intrusive manner, so that the training cost of voice recognition network learning is effectively reduced.

Description

technical field [0001] The embodiments of the present application relate to the field of natural language technology, and in particular to a method, device, device and storage medium for sentence error correction after speech recognition. Background technique [0002] Speech recognition technology is generally implemented using an end-to-end deep learning model. However, once the deep learning model recognizes an error in the text, it will be difficult to manually adjust the model parameters to improve the effect. [0003] The current common practice is to replace the last functional layer of the speech recognition network that decodes into text with a trained language model (Language Model, LM) to assist decoding, so as to improve the text decoding effect. However, the training of other layers of the speech recognition network relies on the trained language model. If the language model is replaced, the entire speech recognition network needs to be retrained. This intrusive ...

AI Technical Summary

Problems solved by technology

However, the training of other layers of the speech recognition network relies on the trained language model. If the language model is replaced, the entire speech recognition network needs to be retrained. This intrusive network design will greatly increase the training cost of speech recognition network learning.

Images