def model_filename(method,
language,
library,
normalization,
accuracy=0,
n_people=None,
n_segments=None,
augmentation=None,
json=True,
models=False):
filename = f'{language}/'
if models:
filename += 'models/'
filename += f'{normalization}/'
filename += verify_people_segments(n_people, n_segments)
filename += verify_augmentation(augmentation)
filename += f'{method}/'
filename += f'{library}/'
if accuracy != 0:
accuracy = Process.pad_accuracy(accuracy)
filename += f'{accuracy}/'
if json:
filename += f'info.json'
return filename
# Audio.write(
# f'portuguese/processed/psf/{dir}_{i}_{sample_index}.wav', sample, rate)
m['attrs'].append(attr.tolist())
del attr
del signal
return m
if __name__ == '__main__':
filename = Directory.processed_filename(language, library, sampling_rate,
n_audios, n_segments, augment)
# m = []
# for j, i in enumerate(f):
# if j < 1:
# m.append(process_directory(i, j, library))
m = Parallel(n_jobs=-1,
verbose=len(f))(delayed(process_directory)(i, j, library)
for j, i in enumerate(f)
if n_audios == None or j < n_audios)
Process.object_to_json(
filename,
m,
f,
)
del m
n_rate = 24000
random_state = 42
filename_ps = Directory.verify_people_segments(people=n_people,
segments=n_segments)
# %%
global X_train, X_valid, X_test, y_train, y_valid, y_test
DATASET_PATH = Directory.processed_filename(language, library, n_rate,
n_people, n_segments)
# %%
# SPLIT DOS DADOS
X_train, X_valid, X_test, y_train, y_valid, y_test = Process.selection(
DATASET_PATH)
mapping = set(y_train)
# %%
def build_model(learning_rate=0.0001):
# build the network architecture
input_shape = [X_train.shape[1]] if library == 'mixed' else [
X_train.shape[1], X_train.shape[2]
]
model = Sequential([
# 1st hidden layer
Flatten(input_shape=input_shape),
Dense(512, activation='relu'),
library = args['representation']
people = args['people']
segments = args['segments']
normalization = args['normalization']
flat = args['flat']
augment = args['augmentation']
sampling_rate = 24000
random_state = 42
epochs = 2000
batch_size = 128
# %%
file_path = Directory.processed_filename(language, library, sampling_rate,
people, segments, augment)
# %%
X_train, X_valid, X_test, y_train, y_valid, y_test = Process.selection(
file_path, flat=flat)
param_grid = {}
# %%
if normalization == 'minmax':
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
X_train = scaler.fit_transform(X_train.reshape(
-1, X_train.shape[-1])).reshape(X_train.shape)
X_test = scaler.transform(X_test.reshape(-1, X_test.shape[-1])).reshape(
X_test.shape)
elif normalization == 'standard':
from sklearn.preprocessing import StandardScaler
language = args['language']
method = 'svm'
library = args['representation']
people = args['people']
segments = args['segments']
normalization = args['normalization']
augment = args['augmentation']
sampling_rate = 24000
random_state = 42
# %%
file_path = Directory.processed_filename(
language, library, sampling_rate, people, segments, augment)
# %%
X_train, y_train, mapping = Process.selection(
file_path, valid_size=0, test_size=0, mapping=True, flat=True)
# %%
if normalization == 'minmax':
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
X_train = scaler.fit_transform(
X_train.reshape(-1, X_train.shape[-1])).reshape(X_train.shape)
elif normalization == 'standard':
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_train = scaler.fit_transform(
X_train.reshape(-1, X_train.shape[-1])).reshape(X_train.shape)
first_folder = Directory.processed_filename('portuguese', 'psf',
sampling_rate, None, None)
second_folder = Directory.processed_filename('portuguese', 'melbanks',
sampling_rate, None, None)
third_folder = Directory.processed_filename('english', 'psf',
sampling_rate, people,
segments)
fourth_folder = Directory.processed_filename('english', 'melbanks',
sampling_rate, people,
segments)
X_train, X_valid, X_test, y_train, y_valid, y_test = Process.mixed_selection(
first_folder,
second_folder,
third_folder,
fourth_folder,
lm_validation=False,
lm_test=False,
rm_validation=True,
rm_test=True)
elif language == 'mixed':
portuguese_folder = Directory.processed_filename('portuguese', library,
sampling_rate, people,
segments)
english_folder = Directory.processed_filename('english', library,
sampling_rate, people,
segments)
X_train, X_valid, X_test, y_train, y_valid, y_test = Process.mixed_selection_language(
portuguese_folder=portuguese_folder,
english_folder=english_folder,