Scene error correction method, device and equipment, and storage medium

Abstract

The embodiment of the invention provides scene error correction method, device and equipment, and a storage medium. The scene error correction method comprises the following steps: performing semanticunderstanding on a training sample by utilizing scene knowledge of a target scene to obtain semantic features; mining associated knowledge of the scene knowledge; and training by utilizing the semantic features, the scene knowledge and the associated knowledge to obtain a scene error correction model, wherein the scene error correction model is used for performing scene error correction on an input text. According to the embodiment of the invention, the scene knowledge is fully utilized to carry out semantic understanding on the training sample, so that finer semantic features can be obtained, and the error correction model effect is improved; and moreover, scene knowledge is fully utilized, and associated knowledge is mined, and an error correction model is assisted to make a decision.

Description

technical field [0001] The present invention relates to the technical field of error correction, in particular to a scene error correction method, device, equipment and storage medium. Background technique [0002] With the emergence of a large number of intelligent voice products, voice interaction has gradually replaced keyboard input as the main interaction mode of intelligent voice products. For example, smart speakers, smart car systems, smart customer service, etc. Speech recognition technology has made breakthrough progress with the blessing of deep learning technology. However, in actual scene speech recognition, due to the lack of semantic understanding and scene knowledge, the error rate of recognition is still relatively high, which seriously affects the user experience of the product. For example, in the map voice scene, the user originally wanted to search for "Yuhuangding Hospital", but due to voice recognition problems, the result of voice recognition was "Y...

AI Technical Summary

Problems solved by technology

However, in the actual scene speech recognition, due to the lack of semantic understanding and scene knowledge, the error rate of recognition is still relatively high, which seriously affects the user experience of the product

For example, in the map voice scene, the user originally wanted to search for "Yuhuangding Hospital", but due to the problem of voice recognition, the result of voice recognition was "Yuhuangding Hospital", which made the user unable to get the result he wanted

For another example, the user wants to know "what time does the yoga room close", but the voice recognition result is "what time does yoga close", which seriously affects the user experience of the entire product

[0003] The main reasons for these errors include: First, the current speech recognition mainly models the speech signal to the text result, which mainly considers the structure of the language itself, and lacks the understanding of semantics

Second, the current speech recognition is mainly aimed at the open domain, and lacks the use of scene knowledge

The error correction model lacks the ability to migrate, and it is difficult to quickly customize it in different error correction scenarios

Images