>
> > 1. LAME jstereo sometimes turns on mid/side stereo when it should not.
> > 2. FhG encoder, which is unquestionably the highest quality, doesn't
> > appear to use jstereo at 160kbs and higher.
>
> So if we want to drop joint-stereo from 160k because of the quality, why
> doing it only from 160k in case of vbr?
> It's obvious that's using cbr, we must use joint-stereo for lower bitrates
> than 160k. But in case of vbr, if we decide to drop joint-stereo because the
> encoder does not always do the best choice, I personnaly think that it's
> quite illogical to drop it only from 160. The encoder sometimes does some
> wrong choices, but it also does them at lower than 160k. So if we want to
> drop joint-stereo in vbr, to be perfectly logical, we must do it for all
> frames, whatever the bitrate. This is quite the same as using the -m s
> switch. So if we got some problems with joint-stereo switching, why not
> doing -m s the default setting for vbr?
>
Okay, it is out! But here are some more thoughts on this, and an
argument for and against:
mid/side stereo encodes the mid and side channels instead of left and
right. It allocates more bits to the mid channel than the side
channel. For signals without a lot of stereo separation, there will
be very little information in the side channel and this trick will
improve bandwidth. If the left & right channels differ by a lot, then
the side channel will contain a lot of information. Errors encoding
this information will show up as noise in *both* the left & right
channels after decoding.
**argument for the 160kbs cutoff**
mid/side stereo will reduce noise in the mid channel and
increase noise in the side channel. lets say the noise
is proportional to the number of bits, and (for example) mid/side
stereo will remove 10db of noise from mid and add 10db of noise
to side. This is generally a good thing since the mid channel
is much stronger than the side channel.
I'm not an audio engineer, so this is probably incorrect, but...
If we are encoding at a high bit rate, with very little noise
created by the encoding, then there may not be 10db of noise to
remove, and thus the 10db reduction in the mid channel will not help.
However, you can always add new noise! Thus the 10db of noise added
to the side channel will still be added, and it will be spread to both
left & right channels and may result in a net increase in total noise.
I think we would need to do lots of tuning to see when mid/side
stereo has a positive effect, and when it has a negative effect.
Fortunately, FhG has probably already done this! It looks like
they decided it is not worth using at 160kbs and higher.
**argument against the 160kbs cutoff**
However, if we put this cutoff into VBR, then we will have a constant
switching between mid/side and regular stereo as the bitrate jumps
between 128 and 160kbs. Since each type of encoding will produce
different noise characteristics, the character of the noise would be
rapidly changing in time. (I've heard this in fixed 128kbs encodings
too - it can be more annoying the the noise itself). I think this is
what the AAC encoders are talking about with their ideas of "temporal
noise shaping". If the shape of the noise changes smoothly in time,
it might me less noticeable?
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
