def training_data_processing(spec_file,
annotation_file,
mean,
std,
spec_file2=None,
annotation_file2=None):
spec = np.load(spec_file)
spec, stretching_rate = pitch_time_deformation_spec(spec)
spec = random_filter_spec(spec)
spec = random_loudness_spec(spec)
label = preprocessing.get_label(annotation_file,
spec.shape[1],
stretching_rate=stretching_rate)
if not (spec_file2 is None):
spec2 = np.load(spec_file2)
spec2, stretching_rate2 = pitch_time_deformation_spec(spec2)
spec2 = random_filter_spec(spec2)
spec2 = random_loudness_spec(spec2)
label2 = preprocessing.get_label(annotation_file2,
spec2.shape[1],
stretching_rate=stretching_rate2)
spec, label = block_mixing_spec(spec, spec2, label, label2)
mels = preprocessing.get_scaled_mel_bands(spec)
mels = preprocessing.normalize(mels, mean, std)
return mels, label
def validation_data_processing(spec_file, annotation_file, mean, std):
spec = np.load(spec_file)
mels = preprocessing.get_scaled_mel_bands(spec)
mels = preprocessing.normalize(mels, mean, std)
n_frame = mels.shape[1]
label = preprocessing.get_label(
annotation_file, n_frame, stretching_rate=1)
return mels, label
def savespec_and_get_bands(file):
audio = utils.load_audio(file)
if len(audio) > 200:
spec = preprocessing.get_spectrogram(audio)
bands = preprocessing.get_scaled_mel_bands(spec)
length = bands.shape[1]
utils.save_matrix(spec, file.replace(".wav", ''))
return length, bands
else:
print("empty file: " + file)
return 0, []
def test_data_processing(spec):
from smd.data import preprocessing
mels = preprocessing.get_scaled_mel_bands(spec)
mels = preprocessing.normalize(mels, mean, std)
return mels.T