Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection
Abstract Unsupervised anomalous sound detection (ASD) aims to detect unknown anomalous sounds of devices when only normal sound data is available. The autoencoder (AE) and self-supervised learning based methods are two mainstream methods. However, the AE-based methods could be limited as the feature...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
SpringerOpen
2023-10-01
|
| Series: | EURASIP Journal on Audio, Speech, and Music Processing |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s13636-023-00308-4 |
| _version_ | 1827709195054481408 |
|---|---|
| author | Jian Guan Youde Liu Qiuqiang Kong Feiyang Xiao Qiaoxi Zhu Jiantong Tian Wenwu Wang |
| author_facet | Jian Guan Youde Liu Qiuqiang Kong Feiyang Xiao Qiaoxi Zhu Jiantong Tian Wenwu Wang |
| author_sort | Jian Guan |
| collection | DOAJ |
| description | Abstract Unsupervised anomalous sound detection (ASD) aims to detect unknown anomalous sounds of devices when only normal sound data is available. The autoencoder (AE) and self-supervised learning based methods are two mainstream methods. However, the AE-based methods could be limited as the feature learned from normal sounds can also fit with anomalous sounds, reducing the ability of the model in detecting anomalies from sound. The self-supervised methods are not always stable and perform differently, even for machines of the same type. In addition, the anomalous sound may be short-lived, making it even harder to distinguish from normal sound. This paper proposes an ID-constrained Transformer-based autoencoder (IDC-TransAE) architecture with weighted anomaly score computation for unsupervised ASD. Machine ID is employed to constrain the latent space of the Transformer-based autoencoder (TransAE) by introducing a simple ID classifier to learn the difference in the distribution for the same machine type and enhance the ability of the model in distinguishing anomalous sound. Moreover, weighted anomaly score computation is introduced to highlight the anomaly scores of anomalous events that only appear for a short time. Experiments performed on DCASE 2020 Challenge Task2 development dataset demonstrate the effectiveness and superiority of our proposed method. |
| first_indexed | 2024-03-10T17:17:30Z |
| format | Article |
| id | doaj.art-c0d3d3ea44da4230abb5d97f0a48aae8 |
| institution | Directory Open Access Journal |
| issn | 1687-4722 |
| language | English |
| last_indexed | 2024-03-10T17:17:30Z |
| publishDate | 2023-10-01 |
| publisher | SpringerOpen |
| record_format | Article |
| series | EURASIP Journal on Audio, Speech, and Music Processing |
| spelling | doaj.art-c0d3d3ea44da4230abb5d97f0a48aae82023-11-20T10:27:35ZengSpringerOpenEURASIP Journal on Audio, Speech, and Music Processing1687-47222023-10-012023111610.1186/s13636-023-00308-4Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detectionJian Guan0Youde Liu1Qiuqiang Kong2Feiyang Xiao3Qiaoxi Zhu4Jiantong Tian5Wenwu Wang6College of Computer Science and Technology, Harbin Engineering UniversityCollege of Computer Science and Technology, Harbin Engineering UniversityBytedanceCollege of Computer Science and Technology, Harbin Engineering UniversityCentre for Audio, Acoustics and Vibration, University of Technology SydneyCollege of Computer Science and Technology, Harbin Engineering UniversityCentre for Vision, Speech and Signal Processing, University of SurreyAbstract Unsupervised anomalous sound detection (ASD) aims to detect unknown anomalous sounds of devices when only normal sound data is available. The autoencoder (AE) and self-supervised learning based methods are two mainstream methods. However, the AE-based methods could be limited as the feature learned from normal sounds can also fit with anomalous sounds, reducing the ability of the model in detecting anomalies from sound. The self-supervised methods are not always stable and perform differently, even for machines of the same type. In addition, the anomalous sound may be short-lived, making it even harder to distinguish from normal sound. This paper proposes an ID-constrained Transformer-based autoencoder (IDC-TransAE) architecture with weighted anomaly score computation for unsupervised ASD. Machine ID is employed to constrain the latent space of the Transformer-based autoencoder (TransAE) by introducing a simple ID classifier to learn the difference in the distribution for the same machine type and enhance the ability of the model in distinguishing anomalous sound. Moreover, weighted anomaly score computation is introduced to highlight the anomaly scores of anomalous events that only appear for a short time. Experiments performed on DCASE 2020 Challenge Task2 development dataset demonstrate the effectiveness and superiority of our proposed method.https://doi.org/10.1186/s13636-023-00308-4Anomalous sound detectionAutoencoderID classifierWeighted anomaly score computation |
| spellingShingle | Jian Guan Youde Liu Qiuqiang Kong Feiyang Xiao Qiaoxi Zhu Jiantong Tian Wenwu Wang Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection EURASIP Journal on Audio, Speech, and Music Processing Anomalous sound detection Autoencoder ID classifier Weighted anomaly score computation |
| title | Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection |
| title_full | Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection |
| title_fullStr | Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection |
| title_full_unstemmed | Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection |
| title_short | Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection |
| title_sort | transformer based autoencoder with id constraint for unsupervised anomalous sound detection |
| topic | Anomalous sound detection Autoencoder ID classifier Weighted anomaly score computation |
| url | https://doi.org/10.1186/s13636-023-00308-4 |
| work_keys_str_mv | AT jianguan transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection AT youdeliu transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection AT qiuqiangkong transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection AT feiyangxiao transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection AT qiaoxizhu transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection AT jiantongtian transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection AT wenwuwang transformerbasedautoencoderwithidconstraintforunsupervisedanomaloussounddetection |