EP3970141B1 - Method and Apparatus for Speech Source Separation Based on a Convolutional Neural Network