deepRnnCnn.py

final_model.py

def final_model(input_dim, filters, kernel_size, conv_stride, 
                conv_border_mode, units, recur_layers, output_dim=29):
    """ Build a deep network for speech 
    """
    # Main acoustic input
    input_data = Input(name='the_input', shape=(None, input_dim))
    # TODO: Specify the layers in your network
    # Add convolutional layer
    conv_1d = Conv1D(filters, kernel_size, 
                     strides=conv_stride, 
                     padding=conv_border_mode,
                     activation='relu',
                     name='conv1d')(input_data)
    pooled_1d = MaxPooling1D(pool_size=2, strides=1)(conv_1d)
    # Add batch normalization
    bn_cnn = BatchNormalization(name='bn_conv_1d')(pooled_1d)
    
    prev_layer = bn_cnn
    for layer in range(recur_layers):
        prev_layer = Bidirectional(GRU(units, activation='relu',
            return_sequences=True, implementation=2, name=f'gru_{layer+1}'))(prev_layer)
        # TODO: Add batch normalization
        prev_layer = BatchNormalization()(prev_layer)
    
    # TODO: Add a TimeDistributed(Dense(output_dim)) layer
    time_dense = TimeDistributed(Dense(output_dim))(prev_layer)
    
    # TODO: Add softmax activation layer
    y_pred = Activation('softmax', name='softmax')(time_dense)
    # Specify the model
    model = Model(inputs=input_data, outputs=y_pred)
    # TODO: Specify model.output_length
    model.output_length = lambda x: cnn_output_length(
        x, kernel_size, conv_border_mode, conv_stride)/2
    print(model.summary())
    return model