Skip to content

Added Keras Layer Wrapper

Bruno Padovese requested to merge augmentation_module_keras_layer_wrapper into augment-on-the-fly

Added the Keras Layer Wrapper to most of the augmentation functions as well as to the magSpectrogram function in spectrogram.py

These layers can now be added via the add_frontend method of the NNArch class.

Have a look at the following example (functioning with the clip NARW waveform data that I created a DB). The spectrogram computation is now done inside the NN in the frontend layer.

frontend = mag.MagnitudeSpectrogramLayer(window_size=256, step_size=32, normalize=True)
nn_interface.model.add_frontend(frontend)

test.py

You can run the script with somethign like this:

python -m test resnet_recipe.json db.h5 waveform.json --train_table /train --val_table /val --output_dir result/

Merge request reports