Training method of pre-training model, encoding feature acquisition method and related device

Abstract

The embodiment of the present disclosure discloses a pre-training model training method, encoding feature acquisition method, and related devices. The method includes: acquiring each audio frame feature and text feature of the audio to be recognized; encoding the text feature through a text encoding module to obtain Text encoding features; randomly select audio frame features for mask processing to obtain masked audio frame feature sequences; through the audio encoding module, combined with text encoding features, encode each masked audio frame feature to obtain masked audio frame encoding feature sequences ;According to the encoding features of each masked audio frame, the features of each training audio frame are obtained, and the audio loss is obtained, and the parameters of the pre-training model are adjusted until the audio loss meets the loss audio threshold, and the trained text encoding module and audio encoding module are obtained. The pre-training model training and encoding feature acquisition method and device disclosed in the embodiments of the present disclosure can improve the accuracy of model training on the basis of a small labeling amount.

Description

technical field [0001] The present disclosure relates to the field of computers, and in particular to a training method for a pre-trained model, a method for acquiring encoding features, and related devices. Background technique [0002] In the field of application and analysis involving voice data, such as teaching analysis, intelligent artificial customer service, automatic language translation, etc., training the data processing model through deep neural network technology, so that it has a strong learning ability and realizes data processing has been more and more popular. The wider the application. [0003] However, the existing data processing model training methods need to train different models based on different applications, which leads to different labeling of data, a large workload of labeling, and high training costs. [0004] Therefore, how to improve the accuracy of model training on the basis of a small amount of labeling has become a technical problem that ...

AI Technical Summary

Problems solved by technology

[0003] However, the existing data processing model training methods need to train different models based on different applications, which leads to different labeling of data, a large workload of labeling, and high training costs

Images