create_dataset.py

from data_processing.mozilla_common_voice import MozillaCommonVoiceDataset
from data_processing.urban_sound_8K import UrbanSound8K
from data_processing.dataset import Dataset
import warnings

warnings.filterwarnings(action='ignore')

mozilla_basepath = '/home/thallessilva/Documents/datasets/en'
urbansound_basepath = '/home/thallessilva/Documents/datasets/UrbanSound8K'

mcv = MozillaCommonVoiceDataset(mozilla_basepath, val_dataset_size=1000)
clean_train_filenames, clean_val_filenames = mcv.get_train_val_filenames()

us8K = UrbanSound8K(urbansound_basepath, val_dataset_size=200)
noise_train_filenames, noise_val_filenames = us8K.get_train_val_filenames()

windowLength = 256
config = {'windowLength': windowLength,
          'overlap': round(0.25 * windowLength),
          'fs': 16000,
          'audio_max_duration': 0.8}

val_dataset = Dataset(clean_val_filenames, noise_val_filenames, **config)
val_dataset.create_tf_record(prefix='val', subset_size=2000)

train_dataset = Dataset(clean_train_filenames, noise_train_filenames, **config)
train_dataset.create_tf_record(prefix='train', subset_size=4000)

## Create Test Set
clean_test_filenames = mcv.get_test_filenames()

noise_test_filenames = us8K.get_test_filenames()

test_dataset = Dataset(clean_test_filenames, noise_test_filenames, **config)
test_dataset.create_tf_record(prefix='test', subset_size=1000, parallel=False)