Machine anomalous sound detection based on a multi-dimensional feature extraction self-encoder model
Wei Dong, Fabin Guo, Tianyao Cheng · 2024
In recent years, machine anomaly sound detection has become a research hotspot. Traditional machine anomaly sound detection tends to use a single audio feature parameter as the feature representation of the sound, and doing so ignores some of the information. In addition, many machine anomaly detection methods draw on image processing to process audio signals, ignoring the temporal nature of audio data. Therefore, this paper proposes an anomaly detection method based on multidimensional feature extraction, which combines Log-Mel spectrogram and Mel-Frequency Cepstral Coefficients features with the aim of extracting more audio information. In this paper, a self-encoder model combining residual convolutional neural network and long and short-term memory network is used to extract features in both spatial and temporal dimensions, respectively, to make full use of the information of the audio signal. Finally, experiments are conducted on anomaly detection dataset to verify the feasibility and effectiveness of the model in this paper.