search

AppleHolic / source_separation

Deep learning based speech source separation using Pytorch
Apache License 2.0
312 stars 45 forks source link

Train using own dataset #23

Closed dmadhitama closed 3 years ago

dmadhitama commented 4 years ago

Hello. Could you please give me an example or explanation about how to train the model using my own dataset? I have some clean, noise, and noisy audio. How to prepare those datasets to be trained in the model? Thank you.

AppleHolic commented 3 years ago

First, If you don't want to fix code, you should make an pandas data frame object to handle Meta Information.

I handled it as some classes (Voice Bank Case Code ), As you can see, it needs json file (meta_path), and it has columns to handle dataset (code).

And, meta files are separated to three files and it needs to saved on same directory code.

If you complete to prepare your own data to fit on above process, you can use meta-directory on train.py.

You can see a sample on Voice Bank preprocessing case - entry point