Deep Recurrent Neural Network-Based Autoencoders for Acoustic Novelty Detection

Figure 2

Block diagram of the proposed acoustic novelty detector with different autoencoder structures. Features are extracted from the input signal and the reconstruction error between the input and the reconstructed features is then processed by a thresholding block which detects the novel or nonnovel event. Structure of the (a) basic autoencoder, (b) compression autoencoder, and (c) denoising autoencoder on the training set or testing set . contains data of nonnovel acoustic events; consists of novel and nonnovel acoustic events.