Singing voice separation using a deep convolutional neural network trained by ideal binary mask and cross entropy