def train_basic_audio_classifier(num_classes, model_savepath, training_data_folder,training_epoch, batch_size = 128, amount_limit=5000, forceLoad = True):
"""
train a basic audio classification model.
"""
epochs = training_epoch
kernel_size = 5
framerate = 16384
(x_train, y_train), (x_test, y_test) = load_audio(training_data_folder, num_classes, forceLoad=forceLoad, framerate=framerate,amount_limit=amount_limit)
print('x_train shape:', x_train.shape)
print('x_test shape:', x_test.shape)
#save_sound(x_train, "classification","xtrain",upscale=False)
#save_sound(x_test, "classification","xtest",upscale=False)
input_shape = (x_train.shape[1],1)
convolution_layers = count_convolutions(input_shape, kernel_size)
model = keras.models.Sequential()
model.add(Conv1D(16, kernel_size=kernel_size, activation="selu", strides=2, input_shape=input_shape, padding="same"))
for i in range(convolution_layers):
model.add(Conv1D(32, kernel_size=kernel_size, activation="selu", strides=2,padding="same"))
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(128, activation="selu"))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(0.0005),
metrics=['accuracy'])
model.summary()
#model.fit(np.expand_dims(x_train, axis=2), y_train,
# batch_size=batch_size,
# epochs=epochs,
# verbose=1,
# validation_data=(np.expand_dims(x_test, axis=2), y_test))
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
model.save(model_savepath)
print('model saved to ', model_savepath)
def build_critic(self):
model = Sequential()
convolution_layers = count_convolutions(self.audio_shape,
self.kernel_size)
input_shape = (self.audio_shape[1], 1)
model = keras.models.Sequential()
model.add(
Conv1D(16,
kernel_size=self.kernel_size,
activation='selu',
strides=2,
input_shape=self.audio_shape,
padding="same"))
for i in range(convolution_layers):
model.add(
Conv1D(32,
kernel_size=self.kernel_size,
activation='selu',
strides=2,
padding="same"))
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(32, activation='selu'))
model.add(Dropout(0.5))
model.add(Dense(1))
model.summary()
clip = Input(shape=self.audio_shape)
validity = model(clip)
return Model(clip, validity)
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
num_classes = 10
(x_train, y_train), (x_test, y_test) = load_audio("speech_commands",
num_classes)
batch_size = 30
epochs = 50
kernel_size = 5
input_shape = (x_train.shape[1], 1)
convolution_layers = count_convolutions(input_shape, kernel_size)
model = keras.models.Sequential()
model.add(Flatten(input_shape=input_shape))
model.add(Dense(512))
model.add(LeakyReLU(alpha=0.2))
model.add(Dense(256))
model.add(LeakyReLU(alpha=0.2))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
model.summary()
model.fit(x_train,
y_train,
idx = np.arange(0 , len(data))
np.random.shuffle(idx)
idx = idx[:num]
data_shuffle = [data[ i] for i in idx]
labels_shuffle = [labels[ i] for i in idx]
return np.asarray(data_shuffle), np.asarray(labels_shuffle)
nClasses=10
(x_train, y_train), (x_test, y_test) = load_audio("speech_commands", nClasses)
kernel_size = 5
sess=tf.InteractiveSession()
convolution_layers = count_convolutions(x_train.shape, kernel_size)
print("{} convolution layers".format(convolution_layers))
#declare input placeholders to which to upload data
tfX=tf.placeholder(dtype=tf.float32,shape=[None,x_train.shape[1],1])
tfY=tf.placeholder(dtype=tf.float32,shape=[None,nClasses])
#build model
layer=tf.layers.conv1d(tfX,16,kernel_size,strides=2,activation=tf.nn.selu)
for i in range(convolution_layers):
layer=tf.layers.conv1d(layer,32,kernel_size,strides=2,activation=tf.nn.selu)
layer=tf.layers.flatten(layer)
layer=tf.layers.dropout(layer,0.5)
layer=tf.layers.dense(layer,32,activation=tf.nn.selu)
layer=tf.layers.dropout(layer,0.5)