Speech classification network training method and device, computing equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a speech classification network training method and device, equipment and a storage medium. The speech classification network training method comprises the following steps: acquiring a small sample data set; taking the training audio samples of the same category as a training set for contrast model learning, pre-training a contrast model based on the training set, and calculating the model loss of the contrast model; iteratively training the comparison model through model loss to obtain a trained comparison model, wherein the trained comparison model comprises a target feature extractor; connecting the target feature extractor with a classifier to construct a speech classification network; and carrying out fine tuning on the speech classification network based on the small sample data set by adopting a small sample learning mode to obtain a trained speech classification network. According to the method, the stability of the data expression ability of the new task is ensured by introducing the supervised learning mode pre-training comparison model, so that the prediction accuracy of the model for the new task is ensured.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a speech classification network training method, device, computing equipment and storage medium. Background technique [0002] At present, with the rapid development of artificial intelligence, speech recognition, as an important part of artificial intelligence, has broad application prospects in different fields. For example, speech emotion recognition has broad application prospects in the fields of human-computer interaction, emotion monitoring or personalized recommendation. The main process of speech emotion recognition includes data preprocessing, feature extraction and emotion classification. However, the current speech emotion recognition often requires a large amount of training data to ensure the accuracy of emotion classification. In practical training, a large amount of label data is difficult to obtain, so meta-learning methods are often used to solve...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present invention provides a speech classification network training method, device, computing equipment and storage medium to solve the problem that the traditional small-sample self-supervised meta-learning method is not stable enough for the data expression ability of the new prediction task and cannot guarantee accurate prediction sexual problems

Images