Vorbis dev - Sep 2009 - How floor1 modifications may affect final (de)coded samples?

Hello,

I've been "playing" a little with the floor1 encoding, and my game
has posed
me a question. I've been modifying (just curiosity) manually, and randomly
(not too great modifications), the quantized [floor1_Y] values, just before
coding them with the codebooks. As, if I'm not wrong, the [floor1_Y]
obtained are then removed to get the residue vectors for each channel, I
(naively) expected the variation introduced manually in the [floor1_Y]
values to be compensated when removing the floor to get the residues.
Although the final modified file sounds quite similar to the original one,
the decoded pcm wave samples are a little bit different (using a simple
Euclidean distance measure, normalized by the number of pcm samples, the
distance is about 0.05 between the original and the modified versions, may
not be a good psycoacoustic distortion measure, but in this game I'm just
interested in the pcm values). Finally, my doubt is: is this caused because
there may not be a quantized residue value to exactly compensate the
variation introduced in the corresponding floor value or am I
misunderstanding something?

Thank you.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/vorbis-dev/attachments/20090917/4ecb1a45/attachment.htm