Voice sample equalization method combining mixed sampling and random forest

Abstract

The invention relates to a voice sample equalization method combining mixed sampling and random forest. The method comprises the following steps: firstly, carrying out feature extraction on an initial voice data set; then carrying out equalization processing on the extracted voice data feature set by using SMOTE-ENN hybrid sampling to obtain a currently equalized voice data set; secondly, inputting the current balanced voice data set into the two-factor random forest model, and outputting a classification evaluation index and an out-of-bag error classification rate of the two-factor random forest model; and finally, judging whether the classification evaluation indexes converge or not, and if the classification evaluation indexes converge, outputting the current balanced voice data set. Otherwise, updating the mixed sampling rate of the SMOTE-ENN mixed sampling according to the out-of-bag error classification rate, returning to carry out equalization processing on the extracted voice data set again until the classification evaluation index converges, and outputting the current equalization voice data set. According to the method, the SMOTE-ENN mixed sampling and the two-factor random forest model are combined to balance the data set, so that the sample data with high information value is reserved to the maximum extent.

Description

technical field [0001] The present invention relates to the technical field of data processing, in particular to a speech sample equalization method, apparatus, device and computer-readable storage medium based on joint mixed sampling and random forest. Background technique [0002] In recent years, artificial intelligence technology has made breakthroughs in speech recognition. However, the problem of data imbalance has always been a challenging problem in machine learning. The data with uneven distribution of categories will cause the recognition ability of the classifier to be significantly biased towards the majority category, and cannot achieve satisfactory classification performance for the minority category. [0003] Currently, traditional imbalanced learning techniques for solving imbalanced data classification problems can be divided into two categories: internal methods and external methods. The internal approach is to improve existing classification algorithms t...

AI Technical Summary

Problems solved by technology

[0005] For this reason, the technical problem to be solved by the present invention is to overcome the SMOTE oversampling in the prior art when generating new samples without considering the distribution of the nearby majority class samples, a lot of noise will be synthesized and invaded into the majority class sample space and ENN undersampling will Defects that cause loss of classification information in datasets

Images