Hi.
I just want to make sure of something which is unclear for me today.
When using MIR-1k dataset, you seem to extract sources from mixture (singing voice+instrumental). Why don't you use a data set where there are mixtures and corresponding sources to load? I don't understand the point of extracting sources from mixture in "preprocess.py" when you want to predict the sources from the mixture by using the network and then compare the prediction and the ground truth.
I hope I'm clear. Correct me if what I wrote above is wrong.
Thanks in advance for your response :)
Hi. I just want to make sure of something which is unclear for me today. When using MIR-1k dataset, you seem to extract sources from mixture (singing voice+instrumental). Why don't you use a data set where there are mixtures and corresponding sources to load? I don't understand the point of extracting sources from mixture in "preprocess.py" when you want to predict the sources from the mixture by using the network and then compare the prediction and the ground truth. I hope I'm clear. Correct me if what I wrote above is wrong. Thanks in advance for your response :)