I could have done that and moved the cutoff frequency back to, oh let's say 4khz but then I'm adding attenuation at the higher end of the band and nonlinear gro

I could have done that and moved the cutoff frequency back to, oh let's say 4khz but then I'm adding attenuation at the higher end of the band and nonlinear group delay. That's maybe fine if that's how we processed the training data, but I don't want us to retrain the classifier every time I change the anti aliasing filter.