def model(flags):
"""Convolutional recurrent neural network (CRNN) model.
It is based on paper
Convolutional Recurrent Neural Networks for Small-Footprint Keyword Spotting
https://arxiv.org/pdf/1703.05390.pdf
Represented as sequence of Conv, RNN/GRU, FC layers.
Model topology is similar with "Hello Edge: Keyword Spotting on
Microcontrollers" https://arxiv.org/pdf/1711.07128.pdf
Args:
flags: data/model parameters
Returns:
Keras model for training
"""
input_audio = tf.keras.layers.Input(shape=modes.get_input_data_shape(
flags, modes.Modes.TRAINING),
batch_size=flags.batch_size)
net = input_audio
if flags.preprocess == 'raw':
# it is a self contained model, user need to feed raw audio only
net = speech_features.SpeechFeatures(
speech_features.SpeechFeatures.get_params(flags))(net)
# expand dims for the next layer 2d conv
net = tf.keras.backend.expand_dims(net)
for filters, kernel_size, activation, dilation_rate, strides in zip(
parse(flags.cnn_filters), parse(flags.cnn_kernel_size),
parse(flags.cnn_act), parse(flags.cnn_dilation_rate),
parse(flags.cnn_strides)):
net = Stream(cell=tf.keras.layers.Conv2D(filters=filters,
kernel_size=kernel_size,
activation=activation,
dilation_rate=dilation_rate,
strides=strides))(net)
shape = net.shape
# input net dimension: [batch, time, feature, channels]
# reshape dimension: [batch, time, feature * channels]
# so that GRU/RNN can process it
net = tf.keras.layers.Reshape((-1, shape[2] * shape[3]))(net)
for units, return_sequences in zip(parse(flags.gru_units),
parse(flags.return_sequences)):
net = GRU(units=units,
return_sequences=return_sequences,
stateful=flags.stateful)(net)
net = Stream(cell=tf.keras.layers.Flatten())(net)
net = tf.keras.layers.Dropout(rate=flags.dropout1)(net)
for units, activation in zip(parse(flags.units1), parse(flags.act1)):
net = tf.keras.layers.Dense(units=units, activation=activation)(net)
net = tf.keras.layers.Dense(units=flags.label_count)(net)
if flags.return_softmax:
net = tf.keras.layers.Activation('softmax')(net)
return tf.keras.Model(input_audio, net)
def model(flags):
"""Gated Recurrent Unit(GRU) model.
It is based on paper
Convolutional Recurrent Neural Networks for Small-Footprint Keyword Spotting
https://arxiv.org/pdf/1703.05390.pdf (with no conv layer)
Hello Edge: Keyword Spotting on Microcontrollers
https://arxiv.org/pdf/1711.07128.pdf
Args:
flags: data/model parameters
Returns:
Keras model for training
"""
input_audio = tf.keras.layers.Input(
shape=(flags.desired_samples,), batch_size=flags.batch_size)
net = speech_features.SpeechFeatures(
frame_size_ms=flags.window_size_ms,
frame_step_ms=flags.window_stride_ms,
sample_rate=flags.sample_rate,
use_tf_fft=flags.use_tf_fft,
preemph=flags.preemph,
window_type=flags.window_type,
mel_num_bins=flags.mel_num_bins,
mel_lower_edge_hertz=flags.mel_lower_edge_hertz,
mel_upper_edge_hertz=flags.mel_upper_edge_hertz,
mel_non_zero_only=flags.mel_non_zero_only,
fft_magnitude_squared=flags.fft_magnitude_squared,
dct_num_features=flags.dct_num_features)(
input_audio)
for units, return_sequences in zip(
parse(flags.gru_units), parse(flags.return_sequences)):
net = GRU(
units=units, return_sequences=return_sequences,
stateful=flags.stateful)(
net)
net = Stream(cell=tf.keras.layers.Flatten())(net)
net = tf.keras.layers.Dropout(rate=flags.dropout1)(net)
for units, activation in zip(parse(flags.units1), parse(flags.act1)):
net = tf.keras.layers.Dense(units=units, activation=activation)(net)
net = tf.keras.layers.Dense(units=flags.label_count)(net)
return tf.keras.Model(input_audio, net)
def model(flags):
"""Gated Recurrent Unit(GRU) model.
It is based on paper
Convolutional Recurrent Neural Networks for Small-Footprint Keyword Spotting
https://arxiv.org/pdf/1703.05390.pdf (with no conv layer)
Model topology is similar with "Hello Edge: Keyword Spotting on
Microcontrollers" https://arxiv.org/pdf/1711.07128.pdf
Args:
flags: data/model parameters
Returns:
Keras model for training
"""
input_audio = tf.keras.layers.Input(shape=modes.get_input_data_shape(
flags, modes.Modes.TRAINING),
batch_size=flags.batch_size)
net = input_audio
if flags.preprocess == 'raw':
# it is a self contained model, user need to feed raw audio only
net = speech_features.SpeechFeatures(
speech_features.SpeechFeatures.get_params(flags))(net)
for units, return_sequences in zip(parse(flags.gru_units),
parse(flags.return_sequences)):
net = GRU(units=units,
return_sequences=return_sequences,
stateful=flags.stateful)(net)
net = Stream(cell=tf.keras.layers.Flatten())(net)
net = tf.keras.layers.Dropout(rate=flags.dropout1)(net)
for units, activation in zip(parse(flags.units1), parse(flags.act1)):
net = tf.keras.layers.Dense(units=units, activation=activation)(net)
net = tf.keras.layers.Dense(units=flags.label_count)(net)
if flags.return_softmax:
net = tf.keras.layers.Activation('softmax')(net)
return tf.keras.Model(input_audio, net)
def model(flags):
"""Gated Recurrent Unit(GRU) model.
It is based on paper
Convolutional Recurrent Neural Networks for Small-Footprint Keyword Spotting
https://arxiv.org/pdf/1703.05390.pdf (with no conv layer)
Hello Edge: Keyword Spotting on Microcontrollers
https://arxiv.org/pdf/1711.07128.pdf
Args:
flags: data/model parameters
Returns:
Keras model for training
"""
input_audio = tf.keras.layers.Input(
shape=(flags.desired_samples,), batch_size=flags.batch_size)
net = speech_features.SpeechFeatures(
speech_features.SpeechFeatures.get_params(flags))(
input_audio)
for units, return_sequences in zip(
parse(flags.gru_units), parse(flags.return_sequences)):
net = GRU(
units=units, return_sequences=return_sequences,
stateful=flags.stateful)(
net)
net = Stream(cell=tf.keras.layers.Flatten())(net)
net = tf.keras.layers.Dropout(rate=flags.dropout1)(net)
for units, activation in zip(parse(flags.units1), parse(flags.act1)):
net = tf.keras.layers.Dense(units=units, activation=activation)(net)
net = tf.keras.layers.Dense(units=flags.label_count)(net)
return tf.keras.Model(input_audio, net)
def model(flags):
"""Convolutional recurrent neural network (CRNN) model.
It is based on paper
Convolutional Recurrent Neural Networks for Small-Footprint Keyword Spotting
https://arxiv.org/pdf/1703.05390.pdf
Represented as sequence of Conv, RNN/GRU, FC layers.
Hello Edge: Keyword Spotting on Microcontrollers
https://arxiv.org/pdf/1711.07128.pdf
Args:
flags: data/model parameters
Returns:
Keras model for training
"""
input_audio = tf.keras.layers.Input(shape=(flags.desired_samples, ),
batch_size=flags.batch_size)
net = speech_features.SpeechFeatures(
frame_size_ms=flags.window_size_ms,
frame_step_ms=flags.window_stride_ms,
sample_rate=flags.sample_rate,
use_tf_fft=flags.use_tf_fft,
preemph=flags.preemph,
window_type=flags.window_type,
feature_type=flags.feature_type,
mel_num_bins=flags.mel_num_bins,
mel_lower_edge_hertz=flags.mel_lower_edge_hertz,
mel_upper_edge_hertz=flags.mel_upper_edge_hertz,
mel_non_zero_only=flags.mel_non_zero_only,
fft_magnitude_squared=flags.fft_magnitude_squared,
dct_num_features=flags.dct_num_features)(input_audio)
# expand dims for the next layer 2d conv
net = tf.keras.backend.expand_dims(net)
for filters, kernel_size, activation, dilation_rate, strides in zip(
parse(flags.cnn_filters), parse(flags.cnn_kernel_size),
parse(flags.cnn_act), parse(flags.cnn_dilation_rate),
parse(flags.cnn_strides)):
net = Stream(cell=tf.keras.layers.Conv2D(filters=filters,
kernel_size=kernel_size,
activation=activation,
dilation_rate=dilation_rate,
strides=strides))(net)
shape = net.shape
# input net dimension: [batch, time, feature, channels]
# reshape dimension: [batch, time, feature * channels]
# so that GRU/RNN can process it
net = tf.keras.layers.Reshape((-1, shape[2] * shape[3]))(net)
for units, return_sequences in zip(parse(flags.gru_units),
parse(flags.return_sequences)):
net = GRU(units=units,
return_sequences=return_sequences,
stateful=flags.stateful)(net)
net = Stream(cell=tf.keras.layers.Flatten())(net)
net = tf.keras.layers.Dropout(rate=flags.dropout1)(net)
for units, activation in zip(parse(flags.units1), parse(flags.act1)):
net = tf.keras.layers.Dense(units=units, activation=activation)(net)
net = tf.keras.layers.Dense(units=flags.label_count)(net)
return tf.keras.Model(input_audio, net)