Unknown Abnormal Sound Recognition of Aircraft Cabin Noise Based on Autoencoder

The acoustic environment inside the aircraft cabin includes various noise sources, such as engine noise, air condition noise, human voices and abnormal noise. Traditional methods are difficult to extract features and identify the abnormal noise in this complex environment. The abnormal noise with known features can be accurately identified based on the trained convolutional neural networks. However, known abnormal noise may also exhibit some unknown characteristics, and many unknown noises and changes may also occur during flight. The existing methods are unable to accurately identify these unknown abnormal sounds. Therefore, this paper proposes an abnormal noise recognition method based on the autoencoder with residual attention network. The autoencoder extracts and reconstructs features from the original sound through encoding and decoding processes. The trained autoencoder has high reconstruction accuracy for the unknown noise. And experiments about the cabin noise were carried out to verify the autoencoder to identify the unknown cabin noise.