mppenc 1.15t is out. It handles teh_sample (http://www.hydrogenaudio.org/forums/index.php?showtopic=30758&hl=teh_sample) properly now. The problem was a result
of aggressive compiler settings in previous versions.
The new versions for both Windows and Linux are now compiled by ak using GCC. It
allows for easier maintenance and ensures that we release compatible binaries
that produce the exact same encoded files.
The new encoder should produce files with a bitrate similar to 1.15s, but in many
cases slightly lower or higher (1-2 kbps at --standard). We've thoroughly tested it
on a Pentium 3, Pentium 4, Athlon, Athlon XP, Athlon 64, G5 (Mac). Speed is mostly
the same as before but on some computers it is slightly lower. Frank Klemm
advises to be very careful when compiling with certain flags and we've
followed this advice.
Future plans:
We're currently testing yet another problematic sample which could be indicative
of a problem with SV7 (SV8 can easily handle it). We're in direct contact with
Frank and it's possible that the next version of the code will include minor
changes to psy.c. It's time to tweak the encoder and slowly finalize SV7
before we tackle the more demanding challenge of a new stream version.
mppenc 1.15t for Windows (http://www.musepack.net/index.php?pg=win)
mppenc 1.15t for Linux (http://www.musepack.net/index.php?pg=lin)
Musepack Tools for Mac OS X (http://www.musepack.net/index.php?pg=osx) (compiled by kuniklo)
Musepack Forum Discussion (http://www.musepack.net/forum/viewtopic.php?t=149)
Nice to see that mpc is alive and problems are taken care of.
I see that the MusePack page says 1.15t "has less problems with highly tonal music compared to 1.14." How safe is it to use 1.15t as a replacement for 1.14?
As safe as letting your mother drive you to work.
There's nothing alphaish in 1.15 anymore. Frank himself said that 1.14 is uninteresting. 1.15t is the natural progression of Klemm's attempts to fix samples such as amnesia, Ventolin, Get Real Paid... In SV8, he managed to improve handling of those. 1.15f was a half-assed attempt, since part of his code that handled those could not be ported back to SV7 (technical limitations of SV7). Our goal is to reach 1.16 and make it as good as possible when it comes to quality. It's a long way, but choosing 1.14 only means going backwards as far as transient-handling is concerned.
As always, trust your ears, not my words.
Great news, thanks for the announcement seed
Very nice, thanks.
Very Good News, Many thanks Seed
Will SV8 become a transform coder or not? I remembered Frank have intended to a transform coder in SV8 somewhere.
It is nice to see that MPC is alive and a team is bringing its development ahead
Seed, when should we await final 1.15 beta?
Seed, when should we await final 1.15 beta?
[a href="index.php?act=findpost&pid=269218"][{POST_SNAPBACK}][/a]
when it's finished
Thanks to the mpc development team !
There will be no final 1.15 beta.
The final beta will be 1.16.
AFAIK the Musepack versioning works like this:
Even numbers -> Beta (eg. 1.14)
Uneven numbers -> Alpha
Edit: Actually Seed explained the development roadmap in his second post.
Even numbers -> Beta (eg. 1.14)
Uneven numbers -> Alpha
Hmm, what about stable then? Seems a bit tricky. ;-)
Even numbers -> Beta (eg. 1.14)
Uneven numbers -> Alpha
Hmm, what about stable then? Seems a bit tricky. ;-)
[a href="index.php?act=findpost&pid=269632"][{POST_SNAPBACK}][/a]
Those are the even numbers that are divisible by an even number of uneven numbers. For dithering reasons.
I couldn’t resist and did few tests yesterday.
1/ creaking.wav
mpc had serious problems with this sample (I’ve often described this artefact as “boiling oil” sounding).
1.14 = the artefact is obvious, and easy to ABX (8/8)
1.15t = the boiling sound is seriously reduced. But additional noise is still audible. Clearly less annoying though. (7/8)
1.14 vs 1.15t = 11 out of 16, pval = 0.105
I was disappointed by this score, because difference seemed to be obvious. I therefore launched a second test: 14 out of 16, pval = 0.002 [/|\ with ABC/HR 1.1b2 you have to shuffle again the samples in order to perform a new test]
TOTAL = 25/32 which should be a significant result (pval = 0.001 according to KikeG’s binomial table)
2/ orion II.wav
another micro-attacks sample
1.14 = there’s additional noise (7/8)
1.15t = noisy too (appeared first as noisier than 1.14) = 8/8
1.14 vs 1.15t = 10 out of 16, pval = 0.227 [pure guess, I didn’t insist]
3/ mahler (chorus)
I’ve used this very tonal sample for my previous ~175 kbps listening test, and it appeared to be the most disappointing one for mpc 1.14.
1.14 = “ugly voices” (7/8)
1.15t = uglier” (8/8)
1.14 vs 1.15t = 10 out of 16, pval = 0.227
I’ve performed a second test to confirm this failure:
12 out of 16, pval = 0.038
This last result is now significant, but the total score isn’t; therefore, last test:
15 out of 16, pval < 0.001
=> TOTAL = 37/48 (pval = 0.000)
I think it could be safe to say that (for my ears) 1.15t lowered the quality of this tonal sample.
4/ Hymnus peregrinorum - Dum pater familias.wav
Another vocal sample, mix of tonal and guttural voices (deep bass). I focused my attention on 2.3 – 4.0 range (raucous voice).
1.14 = sounds distorted (7/8, easy)
1.15t = also disorted, maybe slightly more (7/8)
1.14 vs 1.15t = 12 out of 16, pval = 0.038
It confirmed the (subtle) regression of 1.15t encoding. I must nevertheless add that I was really surprised to obtain a positive score
5/ I went to bed ;-)
log files (uninteresting: no additional comments) (http://www.foobar2000.net/divers/tests/2005.01/mpc115t.7z)
I could upload samples later.
guruboolez what profile were these tests performed on? I believe the most significant 1.15a improvements were on the insane profile
Sorry, I forgot to precise it.
--standard --xlevel for 1.14 & --standard for 1.15t
Could you provide the samples somewhere ? Posting results without any way to check ourselves is a bit useless imo.
If the samples are too long, it would be great if you could also say where exactly the artifact is.
ftp://ftp2.foobar2000.net/foobar/ (http://ftp://ftp2.foobar2000.net/foobar/)
Some are already available. I'll upload the other tomorrow.
Even numbers -> Beta (eg. 1.14)
Uneven numbers -> Alpha
Hmm, what about stable then? Seems a bit tricky. ;-)
Once again, here's how it works:
Alpha = 2 decimal places, odd (e.g. 1.11, 1.13a, 1.15t, etc.)
Beta = 2 decimal places, even (e.g. 1.12, 1.14, etc.)
Release = 1 decimal place (e.g. 1.0, 1.1, etc.)
So, the next logical beta is 1.16, and the next logical release is 1.2.
Regards,
Madrigal
ftp://ftp2.foobar2000.net/foobar/ (http://ftp://ftp2.foobar2000.net/foobar/)
Some are already available. I'll upload the other tomorrow.
[a href="index.php?act=findpost&pid=269691"][{POST_SNAPBACK}][/a]
Thanks
Some samples are very large, like 30s. So could you also define the exact part where you heard the artifacts ?
I was sampling throughout the weekend and didn't find any problems. I do my tests with --insane. I confirmed 15t handling teh_sample. It also seems transaparnet on Bloch_Formule, Walkyries_short, and amnesia. It also tested transparently on Kraftwerk's Tour de France, Eels' My Beloved Monster, and Jolivet Suite en Concert pour flute et percussion - Hardiment.
But, then, I lack the classical range of guruboolez.
I confirmed a problem with the Mahler Chorus, but not in "ugly voices".
14/17 (0.6%) for the first 11 seconds. In this section, what consistently tipped me off was a collapse (or some odd channel variation) of the stereo image toward the end (from about the 8.5 second mark during the TOOOOOOOOOOO).
At the 18-19 second mark, I thought I identified loss in sharpness in the female voice, but at 16/25 (11.5%), it apparently isn't huge (that and I actually preferred the less sharp MPC voice). The male voice in the 20-22 second range seems to have more tonal variation in the original, but I scored 16/25 again.
Since 15s GCC was my preferred version before t (TOOOOO a t, an alpha MPC... ), I tested Mahler on it. 13/17 (2.5%). The problem seemed to have shifted to a little earlier in the section and I felt like I needed to listen more closely to identify which was which. I didn't test the other parts. I did listen to the whole sample to see if 15s GCC created any different problems. Nothing stood out. At 823 KB vs. 878 KB for t, I may stick with 15s GCC. I do want to give the rest of guru's samples a try, however, and will report back later today or tomorrow.
foobar2000 v0.8.3 facilitated the testing.
Hymnus peregrinorum - Dum pater familias
Tested using --insane: transparent
Tried focusing on the section guruboolez indicated along with just before it up to 1.8, the start, and the last 4 seconds. Nothing better than 60%.
samples:
ftp://ftp2.foobar2000.net/foobar/creaking.exe (http://ftp://ftp2.foobar2000.net/foobar/creaking.exe)
ftp://ftp2.foobar2000.net/foobar/Pierres_Reflechies.exe (http://ftp://ftp2.foobar2000.net/foobar/Pierres_Reflechies.exe)
ftp://ftp2.foobar2000.net/foobar/Orion_II.exe (http://ftp://ftp2.foobar2000.net/foobar/Orion_II.exe)
and the old ones:
ftp://ftp2.foobar2000.net/foobar/Hymnus_peregrinorum.zip (http://ftp://ftp2.foobar2000.net/foobar/Hymnus_peregrinorum.zip)
ftp://ftp2.foobar2000.net/foobar/03_mahler.exe.wv (http://ftp://ftp2.foobar2000.net/foobar/03_mahler.exe.wv)
*.exe are wavpack self-extractible files (Win32). Add .wv extension if you want.
Lefungus> for 03_mahler.exe.wv sample, the problem is audible at least on the beginning (first seconds); I've ABXed it on the middle (the loudest part of the sample, near 00:14 - 00:18.