Skip to main content
Topic: Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s (Read 7569 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

I've performed some listening tests between MPC 1.14, 1.15r, 1.15s and the new GCC compile of 1.15s.

I used some killer samples :
  • amnesia (normalized to -1db to avoid clipping).
  • fsol.
  • spahm (cut to avoid clipping, that occurs at the end).

...some difficult samples :
  • astral.
  • transwave.

...an MP3 killer sample :

...and an audiophile sample :
  • rebecca.



I also checked the filesize bug discussed here : http://www.hydrogenaudio.org/forums/index....indpost&p=97966

All encoding were done with the option --xlevel alone, thus at quality 5 (defaulted). They were decoded by Foobar2000 0.8.2, no DSP.

I used ABC/HR. For each sample, I first ran ABX tests. Then ABCHR ratings. With easy samples, I chose to perform 16 ABX sessions. With difficult ones only 8. The decision was made prior each test. I never changed my mind during a test.

Results :

With Amnesia, 1.15r and 1.15s sound bad, 1.14 and 1.15s GCC sound worse.
I ABXed 1.15s versus 1.15s GCC with a score of 8/8.

With Astral, I could ABX 1.15s gcc versus the original. It was the first sample. I don't know if I couldn't ABX the other ones because they sound better or because of listening fatigue.

Badvilbel is transparent for me.

Fsol is horrible whatever version is used.

Rebecca sounds transparent for me.

Spahm sounds worse with 1.15s GCC. I ABXed it versus all other versions.

It seemed to me that Transwave also sounds worse with 1.15s GCC, however I could not ABX it vs 1.15s. 1.15s GCC was the easiest to ABX, though.

The filesize bug is only present in 1.15r. It was solved in 1.15s.

Conclusion :
1.15s GCC is clearly worse than the classical 1.15s on two samples (Amnesia and Spahm). It may be worse with two other samples (Transwave, Astral). It made no difference with the three other samples.

Full results :

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname: MPC115

1R = G:\Hunting bzz\mpc115s\115r\001 Amnesianorm.wav
2L = G:\Hunting bzz\mpc115s\115s\001 Amnesianorm.wav
3L = G:\Hunting bzz\mpc115s\115sGCC\001 Amnesianorm.wav
4L = G:\Hunting bzz\mpc115s\114\001 Amnesianorm.wav

---------------------------------------
General Comments:
In the ABX module, samples 1 and 2 can be immediately recognized because of a hissing. Sample 3 and 4 have more ringing.
---------------------------------------
1R File: G:\Hunting bzz\mpc115s\115r\001 Amnesianorm.wav
1R Rating: 4.0
1R Comment: Hissing at the beginning
---------------------------------------
2L File: G:\Hunting bzz\mpc115s\115s\001 Amnesianorm.wav
2L Rating: 3.0
2L Comment: Hissing at the beginning
---------------------------------------
3L File: G:\Hunting bzz\mpc115s\115sGCC\001 Amnesianorm.wav
3L Rating: 2.0
3L Comment: Hissing at the beginning, ringing
---------------------------------------
4L File: G:\Hunting bzz\mpc115s\114\001 Amnesianorm.wav
4L Rating: 2.0
4L Comment: Hissing at the beginning, ringing
---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\115r\001 Amnesianorm.wav
   17 out of 17, pval < 0.001
Original vs G:\Hunting bzz\mpc115s\115s\001 Amnesianorm.wav
   16 out of 16, pval < 0.001
Original vs G:\Hunting bzz\mpc115s\115sGCC\001 Amnesianorm.wav
   16 out of 16, pval < 0.001
Original vs G:\Hunting bzz\mpc115s\114\001 Amnesianorm.wav
   16 out of 16, pval < 0.001
G:\Hunting bzz\mpc115s\115s\001 Amnesianorm.wav vs G:\Hunting bzz\mpc115s\115sGCC\001 Amnesianorm.wav
   8 out of 8, pval = 0.004

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname: MPC115

1L = G:\Hunting bzz\mpc115s\115sGCC\002 astral.wav
2R = G:\Hunting bzz\mpc115s\115r\002 astral.wav
3L = G:\Hunting bzz\mpc115s\114\001 astral.wav
4R = G:\Hunting bzz\mpc115s\115s\002 astral.wav

---------------------------------------
General Comments:
In the ABX module, the 7 first sessions of sample 1 were done with headphones, spotting sweeping in the third quarter. The last was done with speakers, spotting pumping in the noise.
---------------------------------------
2R File: G:\Hunting bzz\mpc115s\115r\002 astral.wav
2R Rating: 4.0
2R Comment: Sweeping in the 3rd quarter part
---------------------------------------
3L File: G:\Hunting bzz\mpc115s\114\001 astral.wav
3L Rating: 4.0
3L Comment: Sweeping in the 3th quarter part.
---------------------------------------
4R File: G:\Hunting bzz\mpc115s\115s\002 astral.wav
4R Rating: 4.0
4R Comment: Pumped noise
---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\115sGCC\002 astral.wav
   7 out of 8, pval = 0.035
Original vs G:\Hunting bzz\mpc115s\115r\002 astral.wav
   1 out of 1, pval = 0.500

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname: MPC115

1R = G:\Hunting bzz\mpc115s\115s\003 badvilbel.wav
2R = G:\Hunting bzz\mpc115s\115r\003 badvilbel.wav
3R = G:\Hunting bzz\mpc115s\114\001 badvilbel.wav
4R = G:\Hunting bzz\mpc115s\115sGCC\003 badvilbel.wav

---------------------------------------
General Comments:

---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\115r\003 badvilbel.wav
   3 out of 4, pval = 0.313

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname: MPC115

1L = G:\Hunting bzz\mpc115s\115s\004 fsol.wav
2L = G:\Hunting bzz\mpc115s\115sGCC\004 fsol.wav
3R = G:\Hunting bzz\mpc115s\115r\004 fsol.wav
4R = G:\Hunting bzz\mpc115s\114\001 fsol.wav

---------------------------------------
General Comments:
Ugly
---------------------------------------
1L File: G:\Hunting bzz\mpc115s\115s\004 fsol.wav
1L Rating: 1.0
1L Comment:
---------------------------------------
2L File: G:\Hunting bzz\mpc115s\115sGCC\004 fsol.wav
2L Rating: 1.0
2L Comment:
---------------------------------------
3R File: G:\Hunting bzz\mpc115s\115r\004 fsol.wav
3R Rating: 1.0
3R Comment:
---------------------------------------
4R File: G:\Hunting bzz\mpc115s\114\001 fsol.wav
4R Rating: 1.0
4R Comment:
---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\115s\004 fsol.wav
   16 out of 16, pval < 0.001
Original vs G:\Hunting bzz\mpc115s\115sGCC\004 fsol.wav
   16 out of 16, pval < 0.001
Original vs G:\Hunting bzz\mpc115s\115r\004 fsol.wav
   16 out of 16, pval < 0.001
Original vs G:\Hunting bzz\mpc115s\114\001 fsol.wav
   16 out of 16, pval < 0.001

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname: MPC115

1L = G:\Hunting bzz\mpc115s\114\001 Rebecca.wav
2L = G:\Hunting bzz\mpc115s\115r\005 Rebecca.wav
3R = G:\Hunting bzz\mpc115s\115s\005 Rebecca.wav
4L = G:\Hunting bzz\mpc115s\115sGCC\005 Rebecca.wav

---------------------------------------
General Comments:
No problem
---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\114\001 Rebecca.wav
   3 out of 5, pval = 0.500

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname: MPC115

1L = G:\Hunting bzz\mpc115s\115sGCC\007 spahm.wav
2R = G:\Hunting bzz\mpc115s\115s\007 spahm.wav
3L = G:\Hunting bzz\mpc115s\114\001 spahm.wav
4R = G:\Hunting bzz\mpc115s\115r\007 spahm.wav

---------------------------------------
General Comments:

---------------------------------------
1L File: G:\Hunting bzz\mpc115s\115sGCC\007 spahm.wav
1L Rating: 2.0
1L Comment:
---------------------------------------
2R File: G:\Hunting bzz\mpc115s\115s\007 spahm.wav
2R Rating: 4.0
2R Comment:
---------------------------------------
3L File: G:\Hunting bzz\mpc115s\114\001 spahm.wav
3L Rating: 3.0
3L Comment:
---------------------------------------
4R File: G:\Hunting bzz\mpc115s\115r\007 spahm.wav
4R Rating: 4.5
4R Comment:
---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\115sGCC\007 spahm.wav
   8 out of 8, pval = 0.004
Original vs G:\Hunting bzz\mpc115s\115s\007 spahm.wav
   7 out of 8, pval = 0.035
Original vs G:\Hunting bzz\mpc115s\114\001 spahm.wav
   8 out of 8, pval = 0.004
Original vs G:\Hunting bzz\mpc115s\115r\007 spahm.wav
   1 out of 3, pval = 0.875
G:\Hunting bzz\mpc115s\115sGCC\007 spahm.wav vs G:\Hunting bzz\mpc115s\115s\007 spahm.wav
   8 out of 8, pval = 0.004
G:\Hunting bzz\mpc115s\115sGCC\007 spahm.wav vs G:\Hunting bzz\mpc115s\114\001 spahm.wav
   8 out of 8, pval = 0.004
G:\Hunting bzz\mpc115s\115sGCC\007 spahm.wav vs G:\Hunting bzz\mpc115s\115r\007 spahm.wav
   8 out of 8, pval = 0.004

Code: [Select]
ABC/HR Version 0.9b, 30 August 2002
Testname:

1L = G:\Hunting bzz\mpc115s\115s\008 transwave.wav
2R = G:\Hunting bzz\mpc115s\115sGCC\008 transwave.wav
3L = G:\Hunting bzz\mpc115s\114\001 transwave.wav
4R = G:\Hunting bzz\mpc115s\115r\008 transwave.wav

---------------------------------------
General Comments:
Tested from second 11
---------------------------------------
1L File: G:\Hunting bzz\mpc115s\115s\008 transwave.wav
1L Rating: 4.0
1L Comment:
---------------------------------------
2R File: G:\Hunting bzz\mpc115s\115sGCC\008 transwave.wav
2R Rating: 2.0
2R Comment:
---------------------------------------
3L File: G:\Hunting bzz\mpc115s\114\001 transwave.wav
3L Rating: 4.5
3L Comment:
---------------------------------------
ABX Results:
Original vs G:\Hunting bzz\mpc115s\115s\008 transwave.wav
   4 out of 6, pval = 0.344
Original vs G:\Hunting bzz\mpc115s\115sGCC\008 transwave.wav
   7 out of 8, pval = 0.035
G:\Hunting bzz\mpc115s\115s\008 transwave.wav vs G:\Hunting bzz\mpc115s\115sGCC\008 transwave.wav
   0 out of 1, pval = 1.000


Amnesia is from Amnesia - Ibiza (loco mix version), from the Ibiza maxi single (vinyl).
Astral is Astral Projection - Another World (from the Another World CD)
Badvilbel is Autechre - Second Bad Vilbel, from the Anvil Vapre CD
Fsol is The Future Sound of London - Among Myselves, from the Lifeforms CD
Rebecca is Rebecca Pidgeon - Granmother, from the Test CD n°10 of La Nouvelle Revue du Son.
I don't know the origin of Spahm.
Transwave is Transwave - The Rezwalker (London Live Remix), from the Helium CD.

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #1
Ok with all these versions flying around anyone with a definite way to find out what version that you are using? GCC vs Classical wheres the difference other than the filesize?

Pointers to the download points would be appreciated if it has not already been published if it has please ignore my ignorance and point me in the right direction either via PM or a reply.

Interesting results there as well... is this a result of the rounding errors that could of been introduced due to the different compilers? Briefly read the other thread but I don't know much about compilers to understand it completely.

Many regards
-=MusePack... Living Audio Compression=-

Honda - The Power of Dreams

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #2
Thanks for the efforts.
As said already, could you provide links to all the mentionned samples ?
Btw, what do you mean by GCC compile ? We know for sure now "--ffast-math" needs to be disabled. It's not done by default in the src package or in svn yet, so you have to override this manually. If not done already, override everything by just "-02".
It's a 'Jump to Conclusions Mat'. You see, you have this mat, with different CONCLUSIONS written on it that you could JUMP TO.

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #3
Follow-up on some Pio2001 tests.  Didn't bother with 1.14 because that's already dead from previous tests I've done.  Tested spahm, amnesia, and FSOL with --insane preset.  No hiss problem in any of them.  Makes me wonder whether Pio2001 is trying to listen to 32 bit WAVs on a 16 bit card. 

Spahm GCC 2/10
Spahm MSVC 4/10

Amnesia GCC 7/10
Amnesia MSVC 5/10

FSOL GCC 4/10
FSOL MSVC 6/10

For order, the MSVC sample always was 2nd in my tests.  I.e, 1st original vs. GCC.  2nd, original vs. MSVC.  Any physical or mental fatigue from repitition went in its favor.  Amnesia and FSOL I may be able to ABX over a longer test with either version.  With Amnesia, I thought I noticed an additional flaw in the MSVC sample, but turned out to be wrong, so it got a slightly better score on Amnesia as a result of experimentation.  It lost as a result of experimentation on FSOL.

Still, the differences I noted are nothing compared to the dropouts in teh_sample.

The MSVC compile produced the larger file (higher bitrate) for each sample.  The only time that has not been the case was with teh_sample.  To me, that says the GCC compile responds better when it needs more or less bits to encode a sample.  It may turn out that it responds too abruptly.  Right now, however, it seems to be doing it as well as the codec allows.
"All I ask is that composers wash out their ears before they sit down to compose." - Morton Feldman

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #4
Quote
Ok with all these versions flying around anyone with a definite way to find out what version that you are using?


The one linked here : http://www.newmail.ru/messages/file.dhtml?...ppenc-1.15s.zip

Quote
Pointers to the download points would be appreciated if it has not already been published


Sorry, I can't upload the samples in the upload forum. The feature seems to be temporarly broken because of the server move. And my webspace is full.

Quote
Tested spahm, amnesia, and FSOL with --insane preset.  No hiss problem in any of them.  Makes me wonder whether Pio2001 is trying to listen to 32 bit WAVs on a 16 bit card.  [a href="index.php?act=findpost&pid=267690"][{POST_SNAPBACK}][/a]


No, I'm just encoding at quality 5 while you are using quality 7.

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #5
Quote
Pointers to the download points would be appreciated if it has not already been published if it has please ignore my ignorance and point me in the right direction either via PM or a reply.
[{POST_SNAPBACK}][/a]

For future reference, I was able to either locate them via the forum search or, when not available there, via google.  Just type in the sample name plus flac like fsol.flac.  If that doesn't work, try the name of the sample alone or something like fsol samples.

[a href="http://perso.numericable.fr/laguill2/files/fsol.flac]FSOL[/url]
Not sure which Amnesia version Pio2001 used (noramlized or actual).  I tested the normalized version.
Amnesia
The rest you should be able to find on 1 of these 2:
http://ff123.net/samples.html
http://lame.sourceforge.net/download/samples/
"All I ask is that composers wash out their ears before they sit down to compose." - Morton Feldman

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #6
A package with the samples that are not available in ff123.net should be online soon.
I used a new version of Amnesia. I started from one of the 48000 Hz recordings, and instead of normalizing it to 0 db, I normalized it to -1 db.

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #7
Thanks for helping out with testing Pio.

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #8
Quote
Quote
Pointers to the download points would be appreciated if it has not already been published if it has please ignore my ignorance and point me in the right direction either via PM or a reply.
[{POST_SNAPBACK}][/a]

For future reference, I was able to either locate them via the forum search or, when not available there, via google.  Just type in the sample name plus flac like fsol.flac.  If that doesn't work, try the name of the sample alone or something like fsol samples.

[a href="http://perso.numericable.fr/laguill2/files/fsol.flac]FSOL[/url]
Not sure which Amnesia version Pio2001 used (noramlized or actual).  I tested the normalized version.
Amnesia
The rest you should be able to find on 1 of these 2:
http://ff123.net/samples.html
http://lame.sourceforge.net/download/samples/
[a href="index.php?act=findpost&pid=267787"][{POST_SNAPBACK}][/a]


I think AgentMil was talking about download points for the different binaries...
Anyway...  I do!! 

Cheers

Sergio
Sergio
M-Audio Delta AP + Revox B150 + (JBL 4301B | Sennheiser Amperior | Sennheiser HD598)

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #9
Indeed I was...  I downloaded the one from the link and the one from the official MusePack page so which is which? I am confused like a deer staring into two bright lights  Sorry for the n00b questions.


EDIT: The link provided leads to a 404 error...  So I only have the official one from the MusePack site. Which one was better cause if it was the one from the official site then I don't need to hunt for that version. I am happy with whatever was rated better. To wrap it up I don't know which one is GCC and which one is ICC (if thats what the other version used to compiled).


Regards
-=MusePack... Living Audio Compression=-

Honda - The Power of Dreams

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #10
Quote
Indeed I was...  I downloaded the one from the link and the one from the official MusePack page so which is which? I am confused like a deer staring into two bright lights  Sorry for the n00b questions.
[{POST_SNAPBACK}][/a]

MusePack.net is the MSVC compile.  Uncompressed, it's about 80 KB.  The [a href="http://users.tpg.com.au/chirano/mppenc-1.15s.zip]GCC compile[/url] is about 130 KB uncompressed.

GCC md5:  2b0dcf00a06fbf5790b86f9168d2e27c

I prefer GCC.  The answer seems to depend on the preset you select to encode.  If you use insane, use GCC.  If standard, use MSVC.

seed says if you can wait a little (days?), sometime this week they will release a fix that even works when Microsoft compiles it.
"All I ask is that composers wash out their ears before they sit down to compose." - Morton Feldman

Musepack 1.14 vs 1.15r vs old 1.15s vs new 1.15s

Reply #11
Here is a description of the exe that I tested :

Version, display, filesize.

1.14 : MPC Encoder 1.14 -Beta-, 79,360 bytes
1.15r : MPC Encoder 1.15r --Alpha--, 81,408 bytes
1.15s : MPC Encoder 1.15s --Alpha--, 80,384 bytes
1.15s GCC : MPC Encoder 1.15s --Alpha--, 132,096 bytes

I've got the two firsts for years. I got the two 1.15s versions in the news and discussions of this forum.

 
SimplePortal 1.0.0 RC1 © 2008-2019