:: > // pfk
:: > // For 44.1 kHz
:: > // 1... 96 kbps: Mono better than ugly stereo °)
:: > --> // 97...159 kbps: Joint Stereo
:: > --> // 160...192 kbps: Force Joint Stereo bandwidth not enough for LR stereo,
:but reducing switching artefacts
:: > // 193... kbps: Stereo enough bandwidth for LR stereo
:: > //
:: > // °) mostly prevent by automatic downsampling
::
::
:: Why using forced joint by default? I'm personnaly againt it, as it leads to
:: spacialization artifacts.
::
bitrates are calculated for fs=44.1 kHz, scale for other fs
080 096 128 160 192
| | | | |
Lame 3.87 pre-klemm --------------mj------><-----ms--------
my proposal ---mm--><-----mj------><--mf--><--ms---
differences °°°°°°°° ^^^^^^^^
... 96 kbps: stereo makes no sense (for 44.1 kHz), this case is mostly
prevented by automatic downscaling (currently it only occures at 8 and
16 kbps, limit is 17.4 kbps).
Note: * this automatic can be disabled via -mj or -mf or -ms
* this automatic only occures if the MP3 encoder can't lower the
PCM data rate by a further decreasing the fs
* or by forcing high sampling frequencies and low bit rates without
forcing a special stereo mode
97...159 kbps: joint stereo needs the lowest data rate may be causing
switching artefacts.
160...191 kbps: Most music is coded by >95% of MS frames, the resting
5% are not saving so much space. I've not checked the code,
but switching LR <-> MS seems to result in additional bits.
I've found only one piece of music so -mj saves more than 0.1
percent over -mj. Often -mj files are larger.
192... kbps:
-ms as -mf also prevents switching artefacts. If the psycho
accoustic model is correct and the noise shaping is done correctly,
it depends on the correlation coefficent which is the best, r talks
about the degree of the saving (r = +/-1: max, r=0 none)
a high low
data rate channel
+0.5...+2 L+R L-R
+2... oo L R
oo...-2 L R
-2...-0.5 L-R L+R
-0.5...+0.5 R L
"a" may be should not be calculated by the total signal, but by the signal
splitted into several subbands.
So -mf is the best for "a" = +/-0.5...2, -ms for the rest.
Lame pre-klemm forces -ms for >= 160 kbps, so the question is, what
are the problems with -mf if -mf and -ms are nearly the same.
1st problem: weakness of the MS psycho accoustic model (?)
2nd problem: optimum quantization is much more difficult in MS
than in LR, so for a r=0 signal quantization noise
of the MS signal is higher than in the LR model
(for the best process on average both are the same)
3rd problem: music or parts of music with a "a" outside +/-0.5...2
Solutions:
1st problem: I have a lot of ideas, but
* I have to work, lame is only hobby
* days on earth only have 24 h (venus would be nice)
* my english is bad
* literature I have is written in German and is
printed on paper
* explaining difficult problems via email in a foreign
language is very time consumpting (30 min email = 2
min talk). Mostly I take a pencil and a stack of
paper to explain things and this totally fails for
email.
* I can't read a lot of the lame code without having
the tendency to press <Meta>r
2nd problem: May be not so difficult, but CPU time consupting A
rest of <0.5 dB don't plays such a big role.
3rd problem: This is exactly the same problem -ms has with the
complementary. And this happens much much more often.
So if this is a problem -mj is the answer and not -ms
or -mf. But 160 kbps should have enough reservoir
to cover this problem. For -ms I doubt that this is
the case. See ^^^^^^^
Also note that "a" tends to slip from big positive
values directly to opposite negative values.
See also next e-Mail.
--
Mit freundlichen Grüßen
Frank Klemm
eMail | [EMAIL PROTECTED] home: [EMAIL PROTECTED]
phone | +49 (3641) 64-2721 home: +49 (3641) 390545
sMail | R.-Breitscheid-Str. 43, 07747 Jena, Germany
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )