Fig. 2From: Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detectionThe architecture of IDNN uses the reconstruction error of center frame as the anomaly scoreBack to article page