Training method, device, equipment and storage medium for speech classification model

By introducing a variational information bottleneck processing network into the speech classification model, and information compression and iterative training of speech data is solved, the problem of high overfitting and labeling costs in the existing technology is solved, and the classification accuracy and efficiency of the model are improved.

CN114360520BActive Publication Date: 2024-12-24PING AN TECH (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210042806.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-14
Publication Date
2024-12-24
Estimated Expiration
2042-01-14

AI Technical Summary

Technical Problem

Existing speech classification models are prone to overfitting during training, and are limited by the accuracy of the label, sample data volume and sample distribution, and the labeling cost is high.

Method used

The variational information bottleneck processing network is used to compress the sample speech information. By obtaining sample data, preprocessing it into a vector matrix, and inputting it into the variational information bottleneck processing network of the initial speech classification model, the sentence characterization is obtained, and then combined with the sample label pair The model is trained iteratively.

Benefits of technology

It effectively reduces the occurrence of overfitting, improves the classification accuracy of the model, and reduces the dependence on labeling costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114360520B_ABST
    Figure CN114360520B_ABST
Patent Text Reader

Abstract

The present application relates to the field of artificial intelligence, and specifically discloses a training method, device, equipment and storage medium for a speech classification model. The method obtains sample data, wherein the sample data includes sample speech and sample labels corresponding to the sample speech; preprocesses the sample speech to obtain a vector matrix corresponding to the sample speech; inputs the vector matrix into a variational information bottleneck processing network of an initial speech classification model to obtain a sentence representation corresponding to the sample speech; inputs the sentence representation into a classification network of the initial speech classification model to obtain a predicted label, and then iteratively trains the initial speech classification model according to the sample label and the predicted label, thereby obtaining a trained speech classification model. Adding a variational information bottleneck processing network to the speech classification model compresses sample information and suppresses useless information, thereby reducing the occurrence of overfitting.
Need to check novelty before this filing date? Find Prior Art