Method for improving overall classification accuracy of long-tail distribution speech based on transfer learning

Abstract

The invention discloses a method for improving the overall classification accuracy of long-tail distribution speech based on transfer learning, and the method comprises the steps: firstly building an R-CNN model composed of a CNN and an RNN network through the training of a data set presenting long-tail distribution, enabling the CNN network to be used for extracting speech features, and enabling the RNN network to carry out the time sequence modeling of the speech features extracted by the CNN network; further mining speech information, and extracting inter-class separable features for subsequent speech classification; then, training the R-CNN model twice, in the first model training, using data of long tail distribution for model training, and obtaining preliminary model parameters; in the secondary model training, using the data in balanced distribution for model training, and fixing and migrating the CNN network shallow parameters obtained in the primary model training to the secondary model training; and carrying out speech classification prediction by using the model after secondary training, thereby improving the overall classification effect of the speech classification model.

Description

technical field [0001] The invention belongs to the technical field of deep learning network training, in particular to a method for improving the overall classification accuracy of long-tail distribution speech based on transfer learning. Background technique [0002] Speech classification is an important field in the current deep learning field and has a wide range of commercial application value. However, most of the data sets currently used for model training present a long-tailed distribution, which is a special asymmetric distribution. Some of the categories contain a large amount of data, called head categories, and the corresponding other categories contain The amount of data is very small, called tail categories. Since the tail category contains less data than the head category, the classification result will be biased towards the head category, resulting in a deviation of the overall classification result. Often the information contained in the tail category has ...

AI Technical Summary

Problems solved by technology

The difficulty in dealing with long-tail distribution data lies in two aspects, one is the imbalance of data distribution, and the other is the insufficient representation ability of tail categories due to insufficient data volume.

Images