Speech recognition re-scoring method, device and system

CN119028320APending Publication Date: 2024-11-26XG TECHNOLOGIES PTE LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410968663.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The rescoring process in existing speech recognition technologies takes a long time, resulting in low voice recognition efficiency.

Method used

During the speech recognition process, the segmented speech signal is a shorter segment for rescoring, instead of waiting for the entire speech signal to be acquired after the full processing is performed, and the encoded segment and initial recognition segment of each speech segment are rescored.

Benefits of technology

It reduces the time-consuming process of rescoring, improves the efficiency of voice recognition, and enhances the ability to respond quickly to voice signals, especially in application scenarios such as smart driving and smart home.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028320A_ABST
    Figure CN119028320A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a voice recognition re-scoring method, device and system. The method comprises the following steps: acquiring a first voice signal; then based on the first voice signal, determining a first coding segment and a first initial recognition segment of a first voice segment, the duration of the first voice segment being not greater than the duration of the first voice signal; and re-scoring the first coding segment and the first initial recognition segment to determine a re-scoring result of the first voice segment. According to the scheme provided by the embodiment of the invention, during re-scoring, the first coding segment and the first initial recognition segment of the first voice segment are re-scored, and the length of the first voice segment is not greater than that of the first voice signal, namely, the first voice segment is re-scored instead of being re-scored after voice acquisition is finished. And re-scoring is carried out on the obtained coding result and the preliminary recognition result of all the voices, so that the time consumption of the re-scoring process can be reduced, and the voice recognition efficiency is correspondingly improved.
Need to check novelty before this filing date? Find Prior Art