def write(path, data, rate):
import soundfile as sf
from deep_audio import Directory
Directory.create_directory(path, file=True)
sf.write(path, data, rate, subtype='PCM_16')
def plot_cepstrals(data, title=None, x_label='Frame Index', y_label='Index', cmap='magma', size=(10, 6),
caption=None,
fig_name=None,
show=False, close=True):
if size:
plt.figure(figsize=(10, 6), frameon=True)
plt.imshow(data.T,
origin='lower',
aspect='auto',
cmap=cmap,
interpolation='nearest')
plt.title(title)
plt.xlabel(x_label)
plt.ylabel(y_label)
# plt.colorbar(format='%+2.0f')
# plt.clim(vmin, vmax)
if caption:
plt.figtext(0.5, 0.01, caption, wrap=True,
horizontalalignment='center')
if fig_name:
Directory.create_directory(fig_name, True)
plt.savefig(fig_name, transparent=False)
if show:
plt.show()
if close:
plt.close()
def plot(data, title=None, x_label='Time (s)', y_label='Amplitude', size=(10, 6), caption=None,
fig_name=None,
show=False,
close=True):
if size:
plt.figure(figsize=(10, 6), frameon=True)
plt.plot(list(range(0, len(data))), data)
plt.ylabel(y_label)
plt.xlabel(x_label)
plt.title(title)
# Remove a margem no eixo x
plt.margins(x=0)
if caption:
plt.figtext(0.5, 0.01, caption, wrap=True,
horizontalalignment='center')
if fig_name:
Directory.create_directory(fig_name, True)
plt.savefig(fig_name, transparent=False)
if show:
plt.show()
if close:
plt.close()
def plot_subplots(audio, mfccs, lpccs, rate, title=None, size_multiplier=2, cmap='magma', caption=None,
fig_name=None, show=False):
from matplotlib import rcParams, rcParamsDefault
small_size = 8
medium_size = 10
bigger_size = 12
image_size = (10 * size_multiplier, 6 * size_multiplier)
# controls default text sizes
plt.rc('font', size=small_size * size_multiplier)
# fontsize of the axes title
plt.rc('axes', titlesize=small_size * size_multiplier)
# fontsize of the x and y labels
plt.rc('axes', labelsize=medium_size * size_multiplier)
# fontsize of the tick labels
plt.rc('xtick', labelsize=small_size * size_multiplier)
# fontsize of the tick labels
plt.rc('ytick', labelsize=small_size * size_multiplier)
plt.rc('legend', fontsize=small_size *
size_multiplier) # legend fontsize
# fontsize of the figure title
plt.rc('figure', titlesize=bigger_size * size_multiplier)
plt.subplots(2, 2, figsize=image_size)
plt.subplots_adjust(left=0.125, right=0.9, bottom=0.1,
top=0.9, wspace=0.3, hspace=0.3)
plt.suptitle(title)
plt.subplot(2, 2, 1)
plot_audio(audio, rate, close=False, size=None)
plt.subplot(2, 2, 2)
plot_spectrogram(
audio, rate, cmap=cmap, close=False, size=None)
plt.subplot(2, 2, 3)
plot_cepstrals(
data=lpccs, y_label='LPCC Index', cmap=cmap, size=None, close=False)
plt.subplot(2, 2, 4)
plot_cepstrals(
data=mfccs, y_label='MFCC Index', cmap=cmap, size=None, close=False)
if caption:
plt.figtext(0.5, 0.01, caption, wrap=True,
horizontalalignment='center')
if fig_name:
Directory.create_directory(fig_name, True)
plt.savefig(fig_name, transparent=False)
if show:
plt.show()
plt.close()
# Reseta todo o estilo configurado no inicio da função
rcParams.update(rcParamsDefault)
def create_json_file(file, data, indent=2, cls=None):
from deep_audio import Directory
import json
directory = '/'.join(file.split('/')[:-1])
Directory.create_directory(directory)
with open(file, "w") as fp:
json.dump(data, fp, indent=indent, cls=cls)
def plot_spectrogram(data, rate, n_fft=1024, title=None, x_label='Time (s)', y_label='Frequency (kHz)',
cmap='magma', size=(10, 6), caption=None, fig_name=None, show=False, close=True):
if size:
plt.figure(figsize=(10, 6), frameon=True)
plt.specgram(data, NFFT=n_fft, Fs=rate, cmap=cmap)
plt.title(title)
plt.ylabel(y_label)
plt.xlabel(x_label)
if caption:
plt.figtext(0.5, 0.01, caption, wrap=True,
horizontalalignment='center')
if fig_name:
Directory.create_directory(fig_name, True)
plt.savefig(fig_name, transparent=False)
if show:
plt.show()
if close:
plt.close()
metrics=['accuracy'])
return model
# %%
# CRIA O MODELO
learning_rate = 0.0001
model = build_model(learning_rate=learning_rate)
# %%
# SALVA A ESTRUTURA DO MODELO
timestamp = int(time.time())
Directory.create_directory(
f'{language}/models/{model_algo}/{library}/{filename_ps}{timestamp}')
JSON.create_json_file(
f'{language}/models/{model_algo}/{library}/{filename_ps}{timestamp}/model_structure.json',
model.to_json())
model_save_filename = f'{language}/models/{model_algo}/{library}/{filename_ps}{timestamp}/model_weight.h5'
# DECIDE QUANDO PARAR
earlystopping_cb = EarlyStopping(patience=300, restore_best_weights=True)
# SALVA OS PESOS
mdlcheckpoint_cb = ModelCheckpoint(model_save_filename,
monitor="val_accuracy",
save_best_only=True)
model = SVC(C=10, kernel='linear', decision_function_shape='ovo')
filename_holder = Directory.model_filename(
method=method,
language=language,
library=library,
normalization=normalization,
augmentation=augment,
json=False,
models=True
)
model.fit(X_train, y_train)
Directory.create_directory(filename_holder)
dump(model, open(filename_holder + 'model.h5', 'wb'))
dump(scaler, open(filename_holder + 'scaler.pkl', 'wb'))
# SALVA OS PARAMETROS
Model.dump_model(
filename_holder + 'info.json',
language=language,
method=method,
normalization=normalization,
sampling_rate=sampling_rate,
augmentation=augment,
shape=X_train.shape,
seed=random_state,
library=library,
extra={