Anomalous Sound Detection Based on Interpolation Deep Neural Network

Abstract

As the labor force decreases, the demand for labor-saving automatic anomalous sound detection technology that conducts maintenance of industrial equipment has grown. Conventional approaches detect anomalies based on the reconstruction errors of an autoencoder. However, when the target machine sound is non-stationary, a reconstruction error tends to be large independent of an anomaly, and its variations increased because of the difficulty of predicting the edge frames. To solve the issue, we propose an approach to anomalous detection in which the model utilizes multiple frames of a spectrogram whose center frame is removed as an input, and it predicts an interpolation of the removed frame as an output. Rather than predicting the edge frames, the proposed approach makes the reconstruction error consistent with the anomaly. Experimental results showed that the proposed approach achieved 27% improvement based on the standard AUC score, especially against non-stationary machinery sounds.

Multi-Modal Anomaly Detection by Using Audio and Visual Cues

“…In this subsection we compare accuracy of SVM based audio classifier with the deep learning based audio classifier [19]. Datasets used for comparisons are taken from the DCASE Challenge Task 2 [47].…”

“…These 15 datasets are recorded in 15 different environments which are listed in Table 2. Area under the Curve (AUC) scores of [19] and the SVM based technique are compared in Table 2. It can be seen that the performance of both audio classifier is somewhat similar.…”

“…Their solution is based on an auto-encoder, and it uses the residual error, which stands for its reconstruction quality, to identify the anomaly in machine. However, when the target machine sound is nonstationary, a reconstruction error tends to be largely independent of an anomaly, and its variations increase because of the difficulty of predicting the edge frames [19]. This problem is overcome by deep learning based autoregressive networks presented in [20] and interpolation deep neural network presented in [19].…”

“…However, when the target machine sound is nonstationary, a reconstruction error tends to be largely independent of an anomaly, and its variations increase because of the difficulty of predicting the edge frames [19]. This problem is overcome by deep learning based autoregressive networks presented in [20] and interpolation deep neural network presented in [19]. These deep learning based methods achieve high accuracy which is evident from their Area under the Curve (AUC) scores, however, computational complexity of such algorithms is greater than the conventional methods.…”

“…These deep learning based methods achieve high accuracy which is evident from their Area under the Curve (AUC) scores, however, computational complexity of such algorithms is greater than the conventional methods. Furthermore, deep neural network based audio classifiers presented in [20] and [19] have not been tested in surveillance problems. In general, the audio detection is best achieved using SVM with a cost of feature extraction and not having end to end solution.…”