Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improvements of Discriminative Feature Space Training for Anomalous Sound Detection in Unlabeled Conditions

Takuya Fujimura, Ibuki Kuroyanagi, Tomoki Toda

TL;DR

This paper improves the performance of a discriminative method under unlabeled conditions by enhancing the feature extractor to perform better under unlabeled conditions and proposing various pseudo-labeling methods to effectively train the feature extractor.

Abstract

In anomalous sound detection, the discriminative method has demonstrated superior performance. This approach constructs a discriminative feature space through the classification of the meta-information labels for normal sounds. This feature space reflects the differences in machine sounds and effectively captures anomalous sounds. However, its performance significantly degrades when the meta-information labels are missing. In this paper, we improve the performance of a discriminative method under unlabeled conditions by two approaches. First, we enhance the feature extractor to perform better under unlabeled conditions. Our enhanced feature extractor utilizes multi-resolution spectrograms with a new training strategy. Second, we propose various pseudo-labeling methods to effectively train the feature extractor. The experimental evaluations show that the proposed feature extractor and pseudo-labeling methods significantly improve performance under unlabeled conditions.

Improvements of Discriminative Feature Space Training for Anomalous Sound Detection in Unlabeled Conditions

TL;DR

This paper improves the performance of a discriminative method under unlabeled conditions by enhancing the feature extractor to perform better under unlabeled conditions and proposing various pseudo-labeling methods to effectively train the feature extractor.

Abstract

In anomalous sound detection, the discriminative method has demonstrated superior performance. This approach constructs a discriminative feature space through the classification of the meta-information labels for normal sounds. This feature space reflects the differences in machine sounds and effectively captures anomalous sounds. However, its performance significantly degrades when the meta-information labels are missing. In this paper, we improve the performance of a discriminative method under unlabeled conditions by two approaches. First, we enhance the feature extractor to perform better under unlabeled conditions. Our enhanced feature extractor utilizes multi-resolution spectrograms with a new training strategy. Second, we propose various pseudo-labeling methods to effectively train the feature extractor. The experimental evaluations show that the proposed feature extractor and pseudo-labeling methods significantly improve performance under unlabeled conditions.
Paper Structure (17 sections, 5 equations, 2 figures, 3 tables)

This paper contains 17 sections, 5 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. State-of-the-art method under the labeled conditions
  3. Structure of the feature extractor
  4. Training of the feature extractor
  5. Backend
  6. Proposed method
  7. Enhanced feature extractor
  8. Multi-resolution spectrograms
  9. Subspace loss
  10. Pseudo-labeling
  11. Construction of the feature space
  12. Obtaining pseudo-attribute labels by clustering
  13. Experimental evaluations
  14. Experimental setups
  15. Effectiveness of the enhanced feature extractor
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Visualizations of the feature space for the shaker in the source domain of the 2023 dataset, colored by ground-truth labels. The value in the parentheses represents the official score. This is from one of five trials.
  • Figure 2: Visualizations of the feature space for the valve in the source domain of the 2023 dataset, colored by ground-truth labels. The value in the parentheses represents the official score. This is from one of five trials.