Using #libsndfile to read #fft #windows from a #wav file, with a backwards seek to do analysis with #overlap, works great!

However, trying the same trick with an #ogg #vorbis input fails miserably: frame rate drops from 40+ to 3 after some minutes of audio-file time :( I conjecture (based on #top and #strace logs) that it is seeking back to the start of the file and re-decoding the whole thing on every seek backwards a little bit. Unsustainable.

Workaround: #ffmpeg to decompress beforehand.