search

ZFTurbo / MVSEP-CDX23-Cinematic-Sound-Demixing

Model for CDX23 (Cinematic Sound Demixing) contest
38 stars 5 forks source link

asking for the output file too big #2

Open dhhien1503 opened 1 year ago

dhhien1503 commented 1 year ago

Hi there,

Thank you for such a nice code regarding to the cinematic sound demixxing based on your train model which is useful for us. May I ask why the output sound files (5 files) are almost 20 times higher than original file size each.

In the case of training data, can you please guide us how to train the dataset in order to update the training model.

Thank you very much

ZFTurbo commented 1 year ago

1) I store output files as WAV files of type FLOAT, you can change it on 'INT_16' it will have less size.

2) Regarding training I used code from Demucs4 repository. You must organize data the same way they proposed.

dhhien1503 commented 1 year ago

Thanks for your swift reply. I managed to handle it.