While working on comb filters and setting up standalone overlap/merge fft code, I found out that FIR filters need to be doubled (by using (filtfft(filter) ** 2.0). There are thankfully only two such filters to look at.
(Thankfully, FFT filtering on IIR filters as implemented is accurate to within .0005 on 16-bit decoded video!)
While working on comb filters and setting up standalone overlap/merge fft code, I found out that FIR filters need to be doubled (by using (filtfft(filter) ** 2.0). There are thankfully only two such filters to look at.
(Thankfully, FFT filtering on IIR filters as implemented is accurate to within .0005 on 16-bit decoded video!)