| Title |
Experimental evaluation of a relationship between digital filter and autoencoder structures in audio signal-based anomaly detection / |
| Translation of Title |
Ryšio tarp skaitmeninių filtrų ir automatinio kodavimo įrenginių tyrimas, siekiant nustatyti garso signalų pagrįstą anomaliją. |
| Authors |
Tagawa, Yuki |
| Full Text |
|
| Pages |
59 |
| Keywords [eng] |
anomaly detection ; sound data ; machine learning ; unscented Kalman filter ; robust estimation |
| Abstract [eng] |
Sound data-based anomaly detection is getting attraction. This is due to the recent development of a deep neural network applicable for representing complex-structured data, which has propelled its application in a real-world problem. However, the real-world sound data are usually contaminated with background noise, which hinders a model training process for anomaly detection. This report proposed applying various adaptive digital filters for the pre-processing of the sound data and studied the optimization of an autoencoder architecture to improve the anomaly detection performance when noisy data is applied. This study demonstrated the proposed approach with an open-source sound dataset of industrial machinery and discussed the relationship between the adaptive digital filters and the autoencoder architecture. |
| Dissertation Institution |
Kauno technologijos universitetas. |
| Type |
Master thesis |
| Language |
English |
| Publication date |
2021 |
