>
> Xing speed is still a mystery. It's obvious that mmx asm is not the only key
> of their speed...
>
I took a look at some of Mike's Xing/VBR castanets.wav encodings. One
interesting thing: Xing never uses short blocks! So I will go ahead
and guess that they still do not bother with psycho-acoustics. The
probably use reasable, but fixed, masking thresholds (we should add
this to LAME -f) This could explain their speed and pre-echo problems
at 128kbs. If you dont compute the psycho acoustics, you cant tell if
you should do some window switching.
>
> I made a mistake here (I forget the 2 granules for each channel)
> so replace it by:
>
> if (4over<over_limit)
> bitrate--;
>
okay, now I understand. I will add this to LAME. And it is clear
Xing/VBR is doing some very aggressive bitrate reductions. I've seen
them encode some frames as low as 64kbs, while the previous frame was
256kbs.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
