FFT comparison

DFT filters are nifty, but only if they're implemented properly.  This page demonstrates the difference between a well-constructed FFT filter and a subtly broken one, using a real-world example.  For a detailed description of the underlying problem, see my bug report, sent to the Audacity mailing list.

A copy of the test wav file is here.

A small fragment from the capture of Apple II data from an audio cassette:

The wider cycles are 1KHz '1's, while the narrower cycles are 2KHz '0's.  The amplitude of the '1's tends to be much larger.  It's pretty easy to see the low-frequency signal(s) if you sit back far enough.

After conversion by Adobe Audition (FFT filter, high-pass at 600Hz, 256-point FFT with triangular window):

After conversion by Audacity v1.2.3 (FFT filter, high pass at ~600Hz):

Note the Audacity version has unexpected attenuation at the front, plus it retains some parts of the original low-frequency distortion.  As it turns out, Audacity's filter actually adds distortion at around 275Hz.  The bug report linked at the top describes a simple test with a generated 2KHz tone that demonstrates the problem.