search

pfnet-research / meta-tasnet

A PyTorch implementation of Meta-TasNet from "Meta-learning Extractors for Music Source Separation
MIT License
136 stars 16 forks source link

2 stems instead of 4 #8

Open coincoin73 opened 4 years ago

coincoin73 commented 4 years ago

First, thanks for this implementation of improved TasNet model. Now the output produces 4 stems : bass, drum, other and vocal. How can I modify it to achieve 2 stems : accompaniment and vocal only ?

Thanks in advance.

davda54 commented 4 years ago

Hi, thanks for your interest in our work! There's unfortunately not any trivial way of changing the code for 2 stems (it's pretty much hardcoded for musdb), but I believe it shouldn't be that complicated either. Replacing all occurrences of 4 by 2 should basically do the job :) Let me know if you encounter any problems while converting it.

coincoin73 commented 4 years ago

H, I have done the job and it seems to work. I have just tested on very very small dataset to ensure that all is correct. I must do the training on the full musdb18.

davda54 commented 4 years ago

Hi, that's great! Would you consider making a pull request? :)

coincoin73 commented 4 years ago

Hi Impossible to do the complete training as I only have 1 GPU with 16 GB