EP3970141A1 - METHOD AND APPARATUS FOR SPEECH SOURCE SEPARATION BASED ON A CONVOLUTIONAL NEURAL NETWORK