_reorder_sources() in OLA compares scaled/windowed frame with un-scaled/un-windowed frame

asteroid-team / asteroid

The PyTorch-based audio source separation toolkit for researchers

https://asteroid-team.github.io/

MIT License

2.21k stars 419 forks source link

_reorder_sources() in OLA compares scaled/windowed frame with un-scaled/un-windowed frame #688

Closed hendriks73 closed 7 months ago

hendriks73 commented 7 months ago

Hey there,

_reorder_sources() is called with the current and the previous frames as arguments. The way it had been coded, the previous frame might have been scaled or windowed already. But to get best results, we probably want to compare to an unscaled, un-windowed version. That's what this change fixes.

Cheers.

mpariente commented 7 months ago

Hey, Thanks for the contribution.

Look at the diff, because you changed other things (formating), the diff is much less clear. Can you change only the functional part of the code you intend to change please ?

hendriks73 commented 7 months ago

Looked at the diff—I don't think I have changed the formatting. 🤷‍♂️

mpariente commented 7 months ago

You are right, it was functional because of the inplace modification I didn't get.

Merging.