cmds2/run_CnnFeat.sh  --  Generating Convolution Layer Activation and Save to Kaldi Feature Format ---------------------------------------------------------------------------------------------------------------------- Kaldi currently has no components for convolution layers. Thus, we are unable to do decoding with CNN. A workaround for this issue is to generate convolution layer activation with PDNN and save the outputs into Kaldi feature format. Refer to Kaldi+PDNN for more details. This command loads a well-trained CNN model, generates convolution layer activation and save the activation to Kaldi feature format. argument meaning default value --in-scp-file              path to the .scp file for network input features required                     --cnn-param-file CNN parameter file which has been trained by cmds/run_CNN.py required --cnn-cfg-file CNN config file required --out-ark-file path to the (Kaldi-formatted) .ark file to store the network activation required