Language classification method and device, medium and electronic equipment

By combining language identification and automatic speech recognition models, fusing probabilities and adjusting thresholds, the problem of imbalance between accuracy and recall in existing language classification models is solved, achieving more efficient language classification.

CN116386596BActive Publication Date: 2026-03-24BEIJING YOUZHUJU NETWORK TECH CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-01-20
Publication Date
2026-03-24

AI Technical Summary

Technical Problem

Existing language classification models cannot achieve a balance in several aspects at the same time, resulting in poor accuracy and recall, especially when dealing with biased features.

Method used

By combining a language identification model and an automatic speech recognition model, the language of the target audio is determined by fusing the probabilities of the two models. The language probability is configured using the transcription capability of the speech recognition model, and the accuracy and recall are adjusted by setting thresholds.

Benefits of technology

It improves the accuracy and recall of language classification, reduces the impact of biased features on classification, and is suitable for a variety of application scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116386596B_ABST
    Figure CN116386596B_ABST
Patent Text Reader

Abstract

The present disclosure relates to a language classification method, device, medium and electronic equipment. The language classification method comprises: obtaining target audio; inputting the target audio into a trained language recognition model to obtain a first probability that the target audio belongs to each candidate language in a preset candidate language set; inputting the target audio into a trained automatic speech recognition model to obtain a recognition result corresponding to the target audio, wherein the recognition result comprises a language label and a transcription text, and the language label is used to represent a candidate language; configuring a second probability that the target audio belongs to each candidate language according to the recognition result and the transcription ability of the automatic speech recognition model; for each candidate language, fusing the first probability and the second probability corresponding to the candidate language to obtain a fusion probability corresponding to the candidate language; and determining a target language of the target audio according to the fusion probability of all candidate languages and a preset language classification condition, thereby improving the language classification effect.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Language recognition method and device based on neural network and electronic equipment

    CN113380227A