MPC really all it's cracked up to be?

Topic: MPC really all it's cracked up to be? (Read 51206 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

MPC really all it's cracked up to be?

Reply #125 – 2003-07-17 14:40:29

If the issue is soundstage and/or separation, wouldn't it make sense to do the tests without any crossfeed plugins and/or headphone amps that may have crossfeed circuitry?

Edit: I realize that shouldn't affect the sound of the cymbals.

MPC really all it's cracked up to be?

Reply #126 – 2003-07-17 15:18:17

Mppenc 1.15r has been ABXed against 1.14, it's better.

Trying to ABX this sample, you should try MPPENC 1.14 instead of 1.15r.

Trying to solve the problem, you should try 1.15r and 1.95z67 before starting tweaking, maybe this issue is already solved.

MPC really all it's cracked up to be?

Reply #127 – 2003-07-17 15:40:25

Quote

Interesting discussion. Like I said before, since Musepack is a lossy codec it is possible that high-end equipment such as yours could expose flaws in it.

that is all this thread is about and it is most likely not true, it was proven before that hi-tech equipment is not really an issue when testing the lossy codecs, most likely it is either bat-ears or some bork dsp-like piece of hardware imho.

MPC really all it's cracked up to be?

Reply #128 – 2003-07-17 16:45:36

I don't think it is bat ears.

I'm currently running a 16 sessions ABX on it. I'm at the 6th one and I've got 6/6 already. Still 8 rounds to go.

I can't hear any soundstage in this kind of recording, nor in the original, nor in the encoded version.
After about 30 or 50 times listening to 2.wav without noticing anything wrong, I finally become aware of a problem (with the help of encoding at quality 3 first in order to hear what could be wrong). Exactly the same as in the Astral sample : gurgling (is it english ?) sound affecting a faint treble echoey sound to the right. It gurgles most during the drum repetition, and it drop outs (like badvilbel with Lame APS) during the last word ( "YOU" ) of the second sentence of the text.

Don't know if the Senheiser HD600 I got help, that's quite possible, since this sound is subtle. After the test I'll try with some Plantronics headphones.
Well, you can call it "bat ears" in the sense that one must be well trained to MPC artifacts to find it among all the frequencies coming from everywhere at the same time in this recording

Bit exact optical output
Sony DTC 55ES
Arcam Diva A85 ampli
Senheiser HD600 headphones

EDIT : test completed, mppenc 1.14 --quality 5 --xlevel, ABX 14/16

MPC really all it's cracked up to be?

Reply #129 – 2003-07-17 17:20:35

Quote

Edit:
--tmpMask 0 *might* help with the sharpness issue.

Sorry if I haven't been able to express myself clear enough. I guess -tmpMask 0 helps to avoid temporal smearing, pre-echo and similar, but "not as crisp" is rather meant like dull. It sounds somehow like lowpassed, while in the original the high chirping seperates the cymbal from the rest of the "noise" more clearly.

Hopefully I'll be able to repeat the test with your suggestions (and 1.15r) this weekend.

MPC really all it's cracked up to be?

Reply #130 – 2003-07-17 18:08:29

Tried with Plantronics A90 headphones on an SB64 PCI V soundcard.

ABX 7/8, much more difficult, because of poor treble response of the headphones.

MPC really all it's cracked up to be?

Reply #131 – 2003-07-17 18:54:01

Tested Mppenc 1.15r

ABX 8/8

MPC really all it's cracked up to be?

Reply #132 – 2003-07-17 19:21:55

Quote

I finally become aware of a problem (with the help of encoding at quality 3 first in order to hear what could be wrong). Exactly the same as in the Astral sample : gurgling (is it english ?) sound affecting a faint treble echoey sound to the right. It gurgles most during the drum repetition, and it drop outs (like badvilbel with Lame APS) during the last word ( "YOU" ) of the second sentence of the text.

did the same (with q3 first) with 1.15r, abxed q5 --xlevel later with 14/16 B) wow (i used the crappy surfsound speakers again...), so yes it is about training.

edit: gurgling is the word, from dictionary.com:
1. To flow in a broken irregular current with a bubbling sound: water gurgling from a bottle.
2. To make a sound similar to this: The baby gurgled with pleasure.

MPC really all it's cracked up to be?

Reply #133 – 2003-07-17 19:34:37

With the noise level around me
(1st floor, noisy kids outside nearly all the time)
ABXing even fatboy sample on speakers is a no-no.
Not to mention Astral or this.

MPC really all it's cracked up to be?

Reply #134 – 2003-07-17 19:55:39

Quote

Mppenc 1.15r has been ABXed against 1.14, it's better.

Trying to ABX this sample, you should try MPPENC 1.14 instead of 1.15r.

Trying to solve the problem, you should try 1.15r and 1.95z67 before starting tweaking, maybe this issue is already solved.

In some cases. I find one where 1.14 was better (for me) than 1.15r.
http://www.hydrogenaudio.org/forums/index....showtopic=10267
Could some people try with it ? I only obtained one (negative) report.

Before it, I found another one, but can't remember which, and submit it

MPC really all it's cracked up to be?

Reply #135 – 2003-07-17 19:58:58

One second. How did Xerophase pick those samples? Just because he thought they could be difficult to encode + he likes those tunes? Considering all of a sudden so many people could ABX a difference at q5 even with 1.15r with almost randomly selected samples, maybe q5 is not as transparent as we originally thought? Maybe we just didn't know where to concentrate... I think Pio2001's methodology (first listening to q3 and trying to figure out the problems) could probably be applied to other samples too.

MPC really all it's cracked up to be?

Reply #136 – 2003-07-18 01:34:10

I haven't tried with these samples, but Pio's approach works very well for me when trying to ABX most codecs and samples. First you deliberately listen to the samples at a low bitrate/encode condition to hear where things might go wrong, and then you try the target bitrate. It often helps me pick up stuff I might have missed otherwise.

Sort of like a pre-session warm up and familiarisation if you will...

Den.

MPC really all it's cracked up to be?

Reply #137 – 2003-07-18 01:51:35

Refresh my memory, is it harder to encode music from live performances? Somewhere I read that some encoders have problems with live performances. I can't seem to find where I heard this, but it was from either here or way back during the days of the r3mix.net forums.

MPC really all it's cracked up to be?

Reply #138 – 2003-07-18 08:43:45

Quote

I haven't tried with these samples, but Pio's approach works very well for me when trying to ABX most codecs and samples. First you deliberately listen to the samples at a low bitrate/encode condition to hear where things might go wrong, and then you try the target bitrate. It often helps me pick up stuff I might have missed otherwise.

funny thing is i did that with all other encoders (especially vorbis), but somehow i never really tryed lower setting than q5 for mpc as they were usually regarded as not really related to q5. (qualitywise) In any case the problem seems to be unrelated to soft sound or soundstage issues, if i were looking for those the results would have been nill.

MPC really all it's cracked up to be?

Reply #139 – 2003-07-18 08:57:55

The discussion about the forum rules and the policy about blind tests is spitted here :

http://www.hydrogenaudio.org/forums/index....ST&f=25&t=11437

MPC really all it's cracked up to be?

Reply #140 – 2003-07-18 11:04:09

About the soundstage issue: sometimes some people perceive problems in different manners than others. For example, a level mismatch can be perceived as a "cleaner", "more lively", etc sound, not just a diverent level. This could happen too in this case. Maybe there's some problems at high frequencies that Xerophase perceives as reduced soundstage.

I think it could be of some help to know at what MPC quality level the problems go away for Xerophase.

MPC really all it's cracked up to be?

Reply #141 – 2003-07-18 12:59:13

Quote

For example, a level mismatch can be perceived as a "cleaner", "more lively"

could be, but foobar calculates the same track RG for flac and for the mpc in this case. (still could be some parts with different levels i guess)

MPC really all it's cracked up to be?

Reply #142 – 2003-07-18 13:15:31

Quote

Quote
For example, a level mismatch can be perceived as a "cleaner", "more lively"
could be, but foobar calculates the same track RG for flac and for the mpc in this case. (still could be some parts with different levels i guess)

This was just an example, I didn't mean that could be the case now.

MPC really all it's cracked up to be?

Reply #143 – 2003-07-18 13:22:22

Quote

Considering all of a sudden so many people could ABX a difference at q5 even with 1.15r with almost randomly selected samples, maybe q5 is not as transparent as we originally thought?

Well.. it depends. One reason I haven't even bothered with this thread, is that 1.14 was used AND -q5 was used.
I think I've said it before, but if you look enough, it shouldn't be impossible to find something which makes -q5 non-transparent. -q5 is not perfect, it's the standard profile... Imo it's still lots better than anything at that bitrate range.

Everybody acts like there's some new discovery.. I don't know, and I don't have the time to concentrade on this (and I'm not sure how much Frank Klemm has motivation to develope MPC anymore). But I'd be much more interested if these could be ABXed with 1.15r and insane profile..

MPC really all it's cracked up to be?

Reply #144 – 2003-07-18 13:47:54

Quote

Well.. it depends. One reason I haven't even bothered with this thread, is that 1.14 was used AND -q5 was used.
I think I've said it before, but if you look enough, it shouldn't be impossible to find something which makes -q5 non-transparent. -q5 is not perfect, it's the standard profile... Imo it's still lots better than anything at that bitrate range.

Everybody acts like there's some new discovery.. I don't know, and I don't have the time to concentrade on this (and I'm not sure how much Frank Klemm has motivation to develope MPC anymore). But I'd be much more interested if these could be ABXed with 1.15r and insane profile..

Well I don't agree completely with this, because from a development point of view, the most useful killer sample would be one that sounds awful at the standard settings.

Remember, all the cleverness of the codec is in --standard, and the other profiles are basically nothing more than positively and negatively biased (in various ways) versions of it.

If there's an artifact at --standard and you want to raise the bitrate, the reason why it usually won't vanish quickly, is because at any quality level, the psymodel will underestimate the exact same passages in exact same subbands. So when switching quality settings, all improvements will be brought by the thicker margin and lowered curves/thresholds.

So if you find a dramatic defect, the question is not really whether you had enough margin or not, but rather why you did underestimate the ear's capabilities on this very moment (and at these frequency bands).

I hope this is decently clear

MPC really all it's cracked up to be?

Reply #145 – 2003-07-18 13:56:06

No luck ABXing (at work, SB Live & SONY MDR-CD480), have a couple of suggestions:

1) Does --standard --ms 15 make any difference?

2) Is it ABX-able at --insane?

MPC really all it's cracked up to be?

Reply #146 – 2003-07-18 14:15:31

This thread has caused quite some commotion, especially because the starter proved his results.

Quote

and I'm not sure how much Frank Klemm has motivation to develope MPC anymore

This line really got my attention, because I too am wondering what's going to happen to mpc. I mean there's no real mpc portal where all the latest news is announced, this can give 2 impressions :

1. Frank Klemm is working hard to complete SV8 and doesn't feel like releasing new builds
2. mpc development is stagnating for whatever reason

I really hope for point 1 to be true as I've got a lot of albums encoded to mpc, even some I can't get hold off anymore. If this is not the case I really hope mpc's code is released so a developmentcommunity can continue to work on it because I wouldn't want to see mpc dying.

MPC really all it's cracked up to be?

Reply #147 – 2003-07-20 02:26:00

Results for 2.wav :

Mppenc 1.14 --standard --xlevel
160 kbps
ABX 14/16
+7/8 with other headphones = 21/24

Mppenc 1.15r --standard --xlevel
157 kbps
ABX 8/8

Trying to ABX 1.14 vs 1.15r : no way

Mppenc 1.95z67 --standard
159 kbps
ABX 7/8

Trying to ABX 1.95z67 vs 1.15r (1.95z67 seemed worse to me) : 1/8, no way.

Mppenc 1.95z67 --standard --ms 15
171 kbps
ABX 8/8, no audible improvement

Mppenc 1.95z67 --extreme
189 kbps
ABX 7/8 : harder

Trying to ABX --standard vs --extreme : 5/8 : too hard

Mppenc 1.95z67 --insane
215 kbps
ABX given up at 1/4 : no audible difference

_____________

Bit exact SPDIF output
Sony DTC 55ES
Arcam Diva A85
Senheiser HD600

MPC really all it's cracked up to be?

Reply #148 – 2003-07-20 03:00:06

Quote

I really hope for point 1 to be true as I've got a lot of albums encoded to mpc, even some I can't get hold off anymore. If this is not the case I really hope mpc's code is released so a developmentcommunity can continue to work on it because I wouldn't want to see mpc dying.

Me too.

Let's hope for best.

MPC really all it's cracked up to be?

Reply #149 – 2003-07-20 19:31:30

I opposed 1.14 encoder to 1.15r one, at --standard --xlevel setting.
Two parts of 1.wav file :
- 0.0 -> 3.0 sec
- 11.0 -> 14 sec

T E S T - O N E :

Code: [Select]

ABC/HR Version 0.9b, 30 August 2002
Testname: 

1L = C:\Temp\ABX\1.mpc115rQ5.wav
2R = C:\Temp\ABX\1.mpcQ5.wav

---------------------------------------
General Comments:
0.0 - 3.0
ABX test : sample 1 was the worst
---------------------------------------
1L File: C:\Temp\ABX\1.mpc115rQ5.wav
1L Rating: 3.5
1L Comment: 
---------------------------------------
2R File: C:\Temp\ABX\1.mpcQ5.wav
2R Rating: 4.2
2R Comment: 
---------------------------------------
ABX Results:
Original vs C:\Temp\ABX\1.mpc115rQ5.wav
    11 out of 16, pval = 0.105
Original vs C:\Temp\ABX\1.mpcQ5.wav
    10 out of 16, pval = 0.227
C:\Temp\ABX\1.mpc115rQ5.wav vs C:\Temp\ABX\1.mpcQ5.wav
    12 out of 16, pval = 0.038

I had more problems to ABX 1.14. It can be explained by the deafening experience of this sample : more saturated, my hearing doesn't have the same accuracy for the second file. Nevertheless, I previously identified 1.15r to be worst on the ABA part of the test. Last, on direct comparison between two encodings, I founded SAMPLE #1 to be more distorted than SAMPLE#2 (12/16 = 0.04). Therefore, 1.15r have some chances to be worse to my ears than 1.14.
Now, second passage (focused on high power cymbal, which lost some energy by encoding but gained some noise, or kind of grain)

T E S T - T W O :

Code: [Select]

1L = C:\Temp\ABX\1.mpc115rQ5.wav
2R = C:\Temp\ABX\1.mpcQ5.wav

---------------------------------------
General Comments:
11-13

ABX : sample 1 m'a semblé être le pire au début, mais sur la fin ,ce fut l'inverse.
Test à recommencer
---------------------------------------
1L File: C:\Temp\ABX\1.mpc115rQ5.wav
1L Rating: 4.2
1L Comment: 
---------------------------------------
2R File: C:\Temp\ABX\1.mpcQ5.wav
2R Rating: 4.2
2R Comment: 
---------------------------------------
ABX Results:
Original vs C:\Temp\ABX\1.mpc115rQ5.wav
    11 out of 16, pval = 0.105
Original vs C:\Temp\ABX\1.mpcQ5.wav
    7 out of 16, pval = 0.773
C:\Temp\ABX\1.mpc115rQ5.wav vs C:\Temp\ABX\1.mpcQ5.wav
    8 out of 16, pval = 0.598

I couldn't differenciate two encodings during ABA testing. On ABX session, I found SAMPLE#1 to be worse again ; I had good ABX results on the first half, but after the middle of the test, bad results. I deceided to begin again the test, and to do a small pause during ABX session :

T E S T - T W O (bis)

Code: [Select]

1L = C:\Temp\ABX\1.mpcQ5.wav
2L = C:\Temp\ABX\1.mpc115rQ5.wav

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

---------------------------------------
1L File: C:\Temp\ABX\1.mpcQ5.wav
1L Rating: 4.4
1L Comment: 
---------------------------------------
2L File: C:\Temp\ABX\1.mpc115rQ5.wav
2L Rating: 4.0
2L Comment: 
---------------------------------------
ABX Results:
C:\Temp\ABX\1.mpcQ5.wav vs C:\Temp\ABX\1.mpc115rQ5.wav
    13 out of 16, pval = 0.011

No doubts here : ABA and ABX tests are coherent, and I can conclude with good confidence that 1.15r doesn't please my ears. On both passages.

I can't describe with precision my feelings. 1.15r test is more dirty to my ears, more distorted. In comparison, 1.14 seems to be softer, but cleaner. In a analogic way, I'm tempted to compare 1.15r sound to a JPEG or DivX encoding, trying to maintain more details and generating in consequence more noise around edges.
Note that this sample isn't the first to sound worse with 1.15r than with 1.14.

Notice