On Sat, 15 Apr 2000, Ross Levis wrote:
> If anyone is interested,VBR is now more spread over the bitrates. Here is a $
> bitrate. Some -V tuning is obviously needed. I guess the better lossless co$
>
Did you notice any quality differences?
The main difference is Robert's latest VBR settings are now the
defaults. (the old RH_QUALITY_CONTROL and RH_VBR_SIDE switches).
One thing that needs work: The new shortblock huffman encoding saves
50-300 bits per channel per granule, so the VBR code which adds extra
bits for shortblocks probably needs to be re-tuned.
And here's a really esoteric VBR bug:
If VBR uses a large frame (like 320kbs), then ResvMax may become
small, and we have to make sure that main_data_begin doesn't create a
buffer which violates this condition.
The old code wouldn't check the above, but it would pad the current
frame with 0's (wasting bits) to make sure the *next* frame didn't
violate the ResvMax condition. But this results in wasted bits
because in VBR there is a good chance the next frame will be at
a lower bitrate and thus have a larger ResvMax.
Both of these bugs (I think) are now fixed.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
