> Same problem with the sfb21 flag. Should we allow it to be turned
> on and off during VBR encoding based on the number of bits
> being used?
The compression ratio is a good move, but we should also be looking at an
estimate entropy vs bitrate (i.e. in a nearly silent frame 128kbit/s not
11:1 compression of information, it's an expansion!) We should only use MS
when a) we dont think it will cause artifacts and/or b) when the side
channel has less then a threshold of energy determined by a
bitrate/estimated entropy (i.e. lots of free space, then only do MS if
the side is very quiet).
As far as switching on and off features, it would be intresting to see a
spectrogram of the delta between orignal and compressed and see if turing
on and off those features causes visiable discountities.
> > As far as VBR goes would it be wise to ignore the set minimum bit rate and
> > encode at 32Kb/s during frames consisting of nothing but digital silence?
> >
> > This would allow me to encode with a minimum bitrate of 160Kb/s, but allow
> > it to drop to 32Kb/s during frames of true silence and only use M/S if the
> > side channel is almost non-existant.
> >
>
> We should probably do this, but I dont know how much it will
> improve filesize. I'll bet a typical song is about 1% analog silence?
> Meaning the extra code to handle this would save at most 1% in
> MP3 filesize.
People encode more then songs. Consider a good 'book on cd' recording at
64Kbit/s. I havn't done any tests (have to dig out a book on CD disk), but
I imagine that the human speech duty cycle is preety low and dropping to
32Kbit/s would be a good savings. though I wouldn't want to trust the VBR
model not to drop to 32Kbit in the middle of word and cause audiable
distortion.
It would also be intresting to put a weight on differnt bitrates. Right
now w/ VBR we try to find the minimum bitrate that meets our quality
defined by -X n. If you set really high standards, then you'll end up
almost always at the highest bitrates.
I'd like it to something like:
Increase to 224Kb only if distortion at 192K exceets spec by more then 5%
Increase to 256Kb only if distortion at 224K exceeds spec by more then 15%
Increase to 320Kb only if distortion at 256K exceeds spec by more then 35%
This reflect the fact that if 256K wont do it, then 320K probably wont do
it either, and it's not worth the jump in size to goto 320K. Also, it's
possible that the psyco model is confused on a certian frame and is
resulting in our minimums being unresonably low. Basically, at higher
bitrates you get diminishing returns, it would be nice if this were
reflected in the bitrate selection. It would also be good to do the same
for lower bitrates:
Decrease to 128K only if distortion at 160K is less then spec by more then 5%
Decrease to 96K only if distortion at 128K is less then spec by more then 25%
Reflecting my sceptisism that we are actually encoding frames that will
sound as good as spec at those rates, and that the psyco model isn't just
confused. I think this would be preferable to setting a hard minimum
bitrate and would cover my above desire to have a silence->32K setting.
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
