Environmental Sound Classification Based on CNN Latent Subspaces

Mahyub, Maha; Souza, Lincon S.; Batalo, Bojan; Fukui, Kiichi

doi:10.1109/iwaenc53105.2022.9914765

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In general, an autoencoder works with two main functions: the encoder part which functions to compress the input data, and the decoder part which reconstructs the data to its original form. Lastly, another option for audio classification issues is CNN [20][21][22][23]. It was first developed in 1995 by Lecun and Bengio and is very commonly used in image-based classification [24].…”

Section: Introductionmentioning

confidence: 99%

Deep learning and acoustic approach for mechanical failure detection in industrial machinery

Qurthobi,

Maskeliūnas

2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

Research into acoustic signal-based failure detection has developed into a subject that has attracted the attention of many researchers in recent years. Acoustic signal data collection can be performed without having to interrupt or stop the operation of the machine to be inspected. Therefore, it is very beneficial for the development of nondestructive testing and predictive maintenance. In this study, a collection of pump sound recordings that are part of the Malfunctioning Industrial Machine Investigation and Inspection dataset, known as the MIMII dataset, is used as test material. Several deep learning algorithms such as long short-term memory (LSTM), gate recurrent unit (GRU), autoencoder, and convolutional neural network (CNN) were involved and compared to determine their ability to detect failures. Based on the training results with 300 epochs and a learning rate of 10−6 it was found that CNN produced the classification with the highest accuracy compared to the other algorithms. In addition, the CNN algorithm is also capable of performing classification amidst the problem of imbalance in the amount of data.

show abstract