Denoise for Separation
ECS7013P - Deep Learning for Audio and Music final project
Wave-U-Net has shown success in music source separation, but only clean music signals can generate good separation results. In this work, I present a deep denoising autoencoder framework for noisy and reverberant music source separation. I use pretrained Wave-U-Net model to evaluate the source separation performance using denoised audio and the proposed approach shows improvement on the source separation under subjective listening and some objective evaluation metrics.