Fuzzy statement recognition method and system applied to multi-person mixed scene

Abstract

The invention discloses a fuzzy statement recognition method and system applied to a multi-person mixed scene. The method comprises the steps of S1, collecting a voice signal in real time, converting the voice signal into a fuzzy statement text, and preprocessing the fuzzy statement text; S2, performing pinyin-level text error correction on the fuzzy statement text by using an end-to-end pinyin-level text error correction model to obtain a pinyin-level error-free text; S3, performing semantic-level text error detection on the pinyin-level error-free text to obtain the position of an error font; and S4, performing semantic-level correction on the text to obtain a text with clear semantics. According to the method, the text with clear semantics can be recovered from the fuzzy statement text through automatic pinyin-level text error correction and semantic-level text error correction, and the user statement can be recognized more accurately in a multi-person mixed scene.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a fuzzy sentence recognition method and system applied to a scene where multiple people are mixed. Background technique [0002] In natural language processing, fuzzy sentences mainly refer to sentences containing homophonic or near-sounding words, incorrectly used words, and redundant words, which may easily cause the phenomenon that the language processing model cannot be recognized clearly and cannot be processed. The current robot customer service in the business hall is often in a mixed scene where many people are talking at the same time, resulting in the robot customer service not being able to accurately obtain the information that the users who are interacting with it want to express, and the service quality is greatly reduced. Therefore, how to transform fuzzy sentences into semantically accurate sentences is particularly important. Most of the existing...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a fuzzy sentence recognition and system applied in the mixed scene of many people, which mainly solves the problem that the fuzzy sentence cannot be correctly recognized due to various noises in the environment and noisy speech of many people in the mixed scene of many people.

Images