Just a little quality of life tweak — currently, there's no way to pass the desired distance function down the call-stack from freq.STFTLoss (or, by extension, from freq.MultiResolutionSTFTLoss, etc.) to reach STFTMagnitudeLoss.
In the spirit of the change to include w_sc, w_lin_mag, etc. I think it would be helpful to be able to specify this as an optional argument — even if just to select between L1/L2 as already afforded by STFTMagnitudeLoss.
Just a little quality of life tweak — currently, there's no way to pass the desired distance function down the call-stack from
freq.STFTLoss
(or, by extension, fromfreq.MultiResolutionSTFTLoss
, etc.) to reachSTFTMagnitudeLoss
.In the spirit of the change to include
w_sc
,w_lin_mag
, etc. I think it would be helpful to be able to specify this as an optional argument — even if just to select between L1/L2 as already afforded bySTFTMagnitudeLoss
.