Unlock instant, AI-driven research and patent intelligence for your innovation.

Semi-Supervised Learning High Confidence Sample Mining Method for Audio Event Classification

A semi-supervised learning and high-confidence technology, which is applied in the field of semi-supervised learning and high-confidence sample mining, can solve the problems of limited energy for labeling samples and reduce the workload of manual labeling.

Active Publication Date: 2019-02-12
SHANDONG NORMAL UNIV
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Active learning can reduce the workload of manual labeling to a certain extent because it selects samples with high information content for labeling, but it still requires human participation, and in practical applications, the energy of labelers to label samples is limited

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Semi-Supervised Learning High Confidence Sample Mining Method for Audio Event Classification
  • Semi-Supervised Learning High Confidence Sample Mining Method for Audio Event Classification
  • Semi-Supervised Learning High Confidence Sample Mining Method for Audio Event Classification

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0072] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0073] like figure 1 As shown, for those active learning techniques that mine unlabeled audio event samples within the SVM classification boundary, the present invention, after actively learning and labeling a certain number of unlabeled audio event samples, is based on the following three principles for semi-supervised learning. Mining high-confidence samples from within the classification boundary: 1) Smooth hypothesis; 2) Mined positive and negative samples should be as similar as possible to labeled positive and labeled negative samples, respectively; 3) Mined positive and negative samples The positive class samples and negative class samples should be as different as possible from the labeled negative class samples and the labeled positive class samples, respectively. The entire implementation process of the semi-supervised learning high-confidence ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a semi-supervised learning high-confidence sample mining method for audio event classification. The invention innovatively uses three principles to determine the confidence of unlabeled audio event samples, and then mines unlabeled audio event samples with high confidence. Annotate audio event samples. The three principles provide a triple guarantee for the correct labeling of unlabeled audio event samples, and thus can successfully mine unlabeled audio event samples with high confidence for semi-supervised learning. In addition, the three principles of the present invention fully consider the data distribution, and the mined high-confidence samples have a certain diversity, so the classification performance of the audio event classifier can be better improved. The high-confidence samples excavated by the invention are automatically marked and added to the marked audio event sample set, thus improving the classification performance of the classifier without adding additional manual labeling workload, so the invention has a strong practical application application value.

Description

technical field [0001] The invention relates to a semi-supervised learning high-confidence sample mining method for audio event classification. Background technique [0002] Audio event classification refers to the identification of various types of audio events contained in an audio document. Audio event classification is a current research hotspot. A bottleneck problem that restricts the development of audio event classification technology is the problem of sample labeling. Audio event classification usually requires a large number of labeled samples to participate in training during the training phase, and manual sample labeling is very time-consuming and labor-intensive. In some cases, due to too many training samples, it becomes impractical to rely solely on manual labeling. [0003] In order to solve the problem of sample labeling in audio event classification, on the one hand, active learning technology can be used to reduce the workload of manual labeling. Support...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/62
CPCG06F18/2411
Inventor 冷严李登旺方敬程传福万洪林王晶晶
Owner SHANDONG NORMAL UNIV