Figure 3From: Deep neural network-based bottleneck feature and denoising autoencoder-based dereverberation for distant-talking speaker identificationFlowchart of bottleneck feature extraction. In the training stage for a bottleneck DNN, the input of the DNN was a frame, or multi frames, of reverberant speech. The teacher signal was the true speaker labels. For a trained DNN, giving a frame or multiframes of reverberant speech, we took the bottleneck layer as the output feature.Back to article page