Comment by froh

I'd like the article even more if it used numpy.convolve in the benchmark.

comparing pure python and numpy fft feels not right to me.

the result will be similar, sure, but the effect might then only show for much larger inputs.