Well there weren't really many more steps to take. Between FFT and wikipedia article was only PAQ compression, which didn't require prior knowledge. And they had to ignore a few red herrings, but the spectrogram helped there. I wonder how the 30ms rate was ascertained, though - pure guess?
That's exactly what I did -- the spectrogram is from sndfile-spectrogram, not baudline. baudline helpfully draws a line parallel to the frequency axis and shows you what point on the time axis you're at, so measuring times is easy.
