deep-audio-super-resolution

Framework to train audio super-resolution neural nets on spectrogram features

Running an experiment

Append a experiment parameters to the experiments list in models/dnn/experiments.py. e.g.

experiments += [OrderedDict([('dataset', 'speaker1'), ('upsample', 2),('model', 'dnn'), ('phase', 'regression')])]

Options:

dataset: speaker1, multispeaker, music

upsample: 2, 4, 6, 8

model: dnn

phase: cheated, regression

Then prepare the dataset

cd models/dnn/
python prepare_datasets.py

This will create the dataset in a path definted by OUTPUT_DIR and the parameters of the experiment

Then run the experiment

cd models/dnn/
python run_experiments.py

This will save a snapshot of the model in the same path as the data dir with the filename model.snapshot

Generating Samples

cd models/dnn/
python generate_samples.py

This will create the samples in a path definted by OUTPUT_DIR and the parameters of the experiment