IPB

Welcome Guest ( Log In | Register )

3 Pages V   1 2 3 >  
Reply to this topicStart new topic
Problems With Joint Stereo (LAME), high tunes distortion, LAME 3.93
Sonny
post Dec 7 2003, 21:20
Post #1





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



I’ve just registered to share my experience with you. I’ll TRY to be concise.

I’ve experienced some MASSIVE distorsion on high tunes, encoding some Metal songs that require high bitrate to sound good. After MANY try, I know what causes this distorsion : the Joint Stereo (JS) process (I use LAME 3.93 + RazorLame 1.1.5).

JS is supposed to be better than True Stereo (TS). Theorically. Indeed, JS encodes 2 channels, middle (M) and side (S) instead of left (L) and right ( R). Demonstration :
If M=(L+R)/2 and S=(L-R)/2 , then it is easy to get the L and R channel back :
M+S=(L+R+L-R)/2 = 2L/2=L and M-S=(L+R-(L-R))/2=(L+R-L+R)/2=2R/2=R
The trick is : you encode M on one side, and on the other, the more L and R are similar, the less the S channel will need bits to be properly encoded (ex: if L=R (=Mono), the needed space for S channel is zero !)
Conclusion : if you compare 2 files with the same size, the bits gained on similarities between L and R channel on the file encoded with JS enable an overall better encoding quality over the file encoded with TS.

Again, this is theory. Here what happen to me practically :
Sample encoded in 320kbps and TS (archive quality, 1.77Mo). Very close to the WAV sample.
Sample encoded with the --alt-preset standard, in JS (195kbps, 1.07Mo). You can hear the distorsion of high tunes even with a crappy sound board like mine !
Sample encoded with the --alt-preset standard -V 3, in JS (176kbps, 0.98Mo). In this file, I’ve lowered a bit the quality of the encoding (standard is ‘-V 2’). The overall quality should be good but the distorsion is now HUGE. sad.gif
You can imagine what happens if I lower again the quality !!
Now let’s encode the sample in TS.
Sample encoded with the --alt-preset standard -V 4 –m s, in TS (188kbps, 1.04Mo). See? This file was encoded in lower quality (‘only’ -V4), yet, there is no more distortion !! ohmy.gif

Finally, here is a comparison of JS/TS encoding at lower bitrate :
Sample encoded @128kbps with JS (727ko).
Sample encoded @128kbps with TS (727ko).
There is no need to be a sound engineer to get the difference.

To end this test, here is a few remarks:
I’ve tried to encode in JS with the previous version of LAME (3.92), but I got the same result.
I’m not an expert, so tell me where I’m wrong or what could get this thing work good.
I’ve always been a convinced Lame user. I’m not THAT convinced now. But I’m not sure what to do. Continue in TS with Lame or change for something new. Any advice?
Finally, excuse my english, I’m just french!

This post has been edited by Sonny: Dec 7 2003, 21:29
Go to the top of the page
+Quote Post
Hanky
post Dec 7 2003, 21:39
Post #2





Group: Members (Donating)
Posts: 531
Joined: 18-November 01
From: The Netherlands
Member No.: 481



Please post the original losslessly compressed (flac) sample. All users here can do the encoding to mp3 themselves with various settings to do their listening tests.
[edit: Be prepared that forum members are only willing to discuss problems with recommended LAME versions and recommended settings]

This post has been edited by Hanky: Dec 7 2003, 22:20
Go to the top of the page
+Quote Post
frozenspeed
post Dec 7 2003, 22:01
Post #3





Group: Members
Posts: 207
Joined: 16-October 01
From: Seattle, WA
Member No.: 301



just curious, but wasn't this discussed, like, a year ago and there was an update to lame 3.93.1 or something? Or am I way off? =)

Jeff
Go to the top of the page
+Quote Post
saratoga
post Dec 7 2003, 22:22
Post #4





Group: Members
Posts: 4869
Joined: 2-September 02
Member No.: 3264



Wow thats my kind of music. Unfortunately without the orginal file i can't do anything with it. Hopefully you can up a FLAC/APE somewhere?
Go to the top of the page
+Quote Post
Nick Jr III
post Dec 7 2003, 22:23
Post #5





Group: Members
Posts: 278
Joined: 13-January 02
Member No.: 1003



QUOTE (Sonny @ Dec 7 2003, 03:20 PM)
Finally, excuse my english, Iím just french!

Off Topic: -sorry-

you're "just" french ? blink.gif
be proud to be french rolleyes.gif
Go to the top of the page
+Quote Post
Slo Mo Snail
post Dec 7 2003, 23:54
Post #6





Group: Members
Posts: 111
Joined: 2-July 02
From: Germany
Member No.: 2450



Kameloth? biggrin.gif
Yes... fine music smile.gif
Go to the top of the page
+Quote Post
zver
post Dec 8 2003, 00:26
Post #7





Group: Members
Posts: 174
Joined: 12-June 03
From: toronto
Member No.: 7141



I remember lame3.93 had some big problems..Try recomended3.90.3 or 3.93.1 atleast
Go to the top of the page
+Quote Post
Canar
post Dec 8 2003, 00:35
Post #8





Group: Super Moderator
Posts: 3350
Joined: 26-July 02
From: princegeorge.ca
Member No.: 2796



3.90.3 is the recommended version.

3.94betas may be better with some samples; there's been some major tweaking and tuning with that one, but these tweaks and tunes haven't been tested enough to be recommended yet.

If both are broken, report the 3.94beta breakage, along with a lossless copy of the file. (There's the Uploads forum specifically for problems like this.)

We're interested in hearing more about this. Thanks.

Edit: Also, try using --alt-preset extreme or --alt-preset insane. The latter is highly-tuned 320kbps encoding, while the former is highly-tuned ~240kbps(?) encoding. Try both with v3.90.3 and see how it works for you. Dibrom, the tuner, listens to metal as well, and the genre was one of many used in tuning 3.90.3.

This post has been edited by Canar: Dec 8 2003, 00:39


--------------------
You cannot ABX the rustling of jimmies.
No mouse? No problem.
Go to the top of the page
+Quote Post
tigre
post Dec 8 2003, 00:48
Post #9


Moderator


Group: Members
Posts: 1434
Joined: 26-November 02
Member No.: 3890



QUOTE (Canar @ Dec 8 2003, 01:35 AM)
3.94betas may be better with some samples; there's been some major tweaking and tuning with that one, but these tweaks and tunes haven't been tested enough to be recommended yet.

If both are broken, report the 3.94beta breakage, along with a lossless copy of the file. (There's the Uploads forum specifically for problems like this.)

Huh? Who's talked about 3.94betas - have I missed it?


Anyway - Sonny: It has been said before ... for others being able to verify your findings it's necessary to upload a losslessly compressed (flac, monkey's, wavpack ...) sample. You can use upload forum for this and put a link to the upload thread here.

Have you mp3gained your files or used replaygain on playback to avoid clipping? If not, this could be a reason for the problems you hear too (Only a possibility - I haven't listened to your samples as without knowing the original this would be quite useless, so I can't tell for sure).


--------------------
Let's suppose that rain washes out a picnic. Who is feeling negative? The rain? Or YOU? What's causing the negative feeling? The rain or your reaction? - Anthony De Mello
Go to the top of the page
+Quote Post
Sonny
post Dec 8 2003, 01:07
Post #10





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



I made a small mistake : after checking again, I didn't use lame 3.93 but lame 3.93.1 for this test. I've also tried the 3.92 version (same problems), but not the 3.90.

And all setting are from the recommanded Lame setting list (alt preset standard and alt preset cbr). I just changed the '-V' parameter on sample 2 and 3.
I've managed to save some spaces on my ftp so here is the LINK REMOVED. cool.gif

I also forgot something : I've downloaded a CBR 128kbps mp3 of this song that I've found on the net. It was encoded with FhG and Joint Stereo. Well, I was surprised to hear that it has the same noisy distortion. Compares to my two others 128kbps samples, it sounds better than the JS one (less distortion) but worse than the True Stereo.

So I may conclude that for this kind of music -hungry for bits, with high tunes- the True Stereo compression sounds better than the Joint Stereo.
That's completly against what I've always read about... unsure.gif

Oh and btw, yes, Kamelot is good music ! biggrin.gif

Edit:
Sonny, the forum rules say
QUOTE
Fair use of short music clips (under 30 seconds) for codec testing purposes are allowed.
Please respect them (and read them before posting if you haven't done yet). For verifying your findings a sample of 20-30 seconds should be more than enough if it's as obvious as you say. Thank-you.

edit2: I'll attatch a cut sample (30 seconds) to this post, if it doesn't contain the positions you're talking about, upload your own.


This post has been edited by tigre: Dec 8 2003, 10:56
Attached File(s)
Attached File  test_30sec.ape ( 0bytes ) Number of downloads: 20
 
Go to the top of the page
+Quote Post
Sonny
post Dec 8 2003, 01:34
Post #11





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



QUOTE (Canar @ Dec 7 2003, 03:35 PM)
3.90.3 is the recommended version.

Edit: Also, try using --alt-preset extreme or --alt-preset insane. The latter is highly-tuned 320kbps encoding, while the former is highly-tuned ~240kbps(?) encoding. Try both with v3.90.3 and see how it works for you. Dibrom, the tuner, listens to metal as well, and the genre was one of many used in tuning 3.90.3.

I'll try with the 3.90.3 version and let you know. For now I stick with the True Stereo compression. dry.gif

And yes, I've also tried the alt preset extreme (but not the 'insane'). I could still hear this crappy distortion though the sample sounded a bit better than with the standard preset. Logical, given it's encoded at higher bitrates (204kbps).

When you boost up the encoding quality with the '-V' parameter, the distortion is less and less audible but, hey, what's the point of encoding with VBR at nearly 240kbps ('extreme -V0') when you reach the same result at 200kbps or less with the TS (still talking of that specific distortion) ?!?!
Go to the top of the page
+Quote Post
kjempen
post Dec 8 2003, 06:10
Post #12





Group: Members
Posts: 137
Joined: 3-October 01
Member No.: 193



Sonny:

If you want people to take you more seriously in this forum, you should:

a. provide a losslessly compressed sample (FLAC or something like that)

b. provide some ABX results. Just saying this or that sounds better doesn't really prove anything without doing the proper testing

c. steer clear of doing too much command line tweaking unless you really know what you are doing, and are very familiar with the inner workings of LAME. (Usually you end up with worse results after "tweaking".)
Go to the top of the page
+Quote Post
JohnV
post Dec 8 2003, 06:21
Post #13





Group: Developer
Posts: 2797
Joined: 22-September 01
Member No.: 6



I haven't had time to test this, but it's pure madness to decide something based on one sample, especially when lots of testing has shown that mid/side coding is very clearly useful in alt-presets.
And this is lossy audio - of course there are exceptions. But the author should at least provide some blind listening test results first.


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
Hanky
post Dec 8 2003, 10:22
Post #14





Group: Members (Donating)
Posts: 531
Joined: 18-November 01
From: The Netherlands
Member No.: 481



I downloaded the original sample. Encoded with --aps (LAME 3.90.3)
A listening test between wav and the encoded file did not show any problems here.
[edit:]
westgroveg is right: retested with 3.93.1 , No (at least no obvious) problems either

This post has been edited by Hanky: Dec 8 2003, 11:28
Go to the top of the page
+Quote Post
tigre
post Dec 8 2003, 10:40
Post #15


Moderator


Group: Members
Posts: 1434
Joined: 26-November 02
Member No.: 3890



Sonny, I tested with the (too long, see edit above) .ape sample you provided: lame 3.90.3 --aps, --aps -m s and couldn't hear anything obviously wrong. So please ...

- tell where exactly the problem is most obvious and what exactly it sounds like. Then upload an .ape sample < 30 seconds containing this position.

- perform an ABX test to prove that it's not just immagination using e.g. foobar2000's ABX comparator (other ABX tools: see the FAQ

Besides: There's indeed clipping on decoding if no replaygain/mp3gain is applied. It doesn't cause to me obvious problems but might be a reason for what you hear. So please use mp3gain or replaygain for testing (fb2k's ABX comparator forces you to use replaygain anyway ...).

Please try to understand. It has happened quite often that people come here, claim that there's something terribly wrong with some codec, but when they're asked to support their statement by providing double blind test results (ABX) and a sample they suddenly disappeared. Things like this cause the forums being polluted with wrong and/or unimportant information making valuable information hard to find. Also, audible problems reported by codec users can only be useful for developers (many of them read the forums here) if there's evidence. Some of the forum rules are there because of this.

As far as I understand your theory about "true" and joint stereo correctly, there's a basic mistake: The joint stereo used by the -alt-presets switch between "true" stereo (=Left/Right) and Middle/Side automatically, depending on the redundancy between channels. In many cases this saves bits that can be used to avoid audible problems. (Details see FAQ.)


--------------------
Let's suppose that rain washes out a picnic. Who is feeling negative? The rain? Or YOU? What's causing the negative feeling? The rain or your reaction? - Anthony De Mello
Go to the top of the page
+Quote Post
Sonny
post Dec 8 2003, 11:12
Post #16





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



QUOTE (kjempen @ Dec 7 2003, 09:10 PM)
Sonny:

If you want people to take you more seriously in this forum, you should:

a. provide a losslessly compressed sample (FLAC or something like that)

b. provide some ABX results. Just saying this or that sounds better doesn't really prove anything without doing the proper testing

c. steer clear of doing too much command line tweaking unless you really know what you are doing, and are very familiar with the inner workings of LAME. (Usually you end up with worse results after "tweaking".)

a. READ THE PREVIOUS POSTS BEFORE REPLYING

b. I trust my ears

c. I didn't know that the 'alt preset standard', 'alt preset cbr' and 'alt preset extreme' was "tweaked" command line.
Go to the top of the page
+Quote Post
westgroveg
post Dec 8 2003, 11:12
Post #17





Group: Members
Posts: 1235
Joined: 5-October 01
Member No.: 220



QUOTE (Sonny @ Dec 8 2003, 10:01 PM)
QUOTE (outscape @ Dec 7 2003, 06:28 PM)
why? in most cases joint stereo is better than stereo.

uh wot the hell ? can you read the previous post BEFORE replying ?

People try 3.93.1. Isn't hydrogenaudio interested in testing new problem samples anymore?
Go to the top of the page
+Quote Post
Nick Jr III
post Dec 8 2003, 11:23
Post #18





Group: Members
Posts: 278
Joined: 13-January 02
Member No.: 1003



QUOTE (Sonny @ Dec 8 2003, 05:12 AM)
b. I trust my ears

dry.gif
hem...

You'd better to use ABX to get OBJECTIVE results...

I've made this mistake too to trust my ears and the conclusions show it's vey easy to suffer for the placebo effect.
Mea culpa. laugh.gif

Regards to all HA members thanks to which I understood. wink.gif

This post has been edited by Nick Jr III: Dec 8 2003, 11:29
Go to the top of the page
+Quote Post
Moneo
post Dec 8 2003, 11:27
Post #19





Group: Developer
Posts: 501
Joined: 22-January 03
From: Netherlands
Member No.: 4684



QUOTE (Sonny @ Dec 8 2003, 11:12 AM)
b. I trust my ears

This argument can not, and will not, be accepted.

Time and again, sighted evaluation of audio quality has been proved to be unreliable. This is the reason why in this community statements regrding sound quality have to be supported by blind test results.

Please follow the suggestions given above: use replaygain and do an ABX test if you want your claims to be considered as valid.
Go to the top of the page
+Quote Post
tigre
post Dec 8 2003, 11:33
Post #20


Moderator


Group: Members
Posts: 1434
Joined: 26-November 02
Member No.: 3890



QUOTE (westgroveg @ Dec 8 2003, 12:12 PM)
People try 3.93.1. Isn't hydrogenaudio interested in testing new problem samples anymore?

How do you come to this conclusion? If you browse "I found a problem ..." threads you'll find 3 types:

1. ppl knowing and respecting the rules = providing ABX results + uploading samples
=> responses showing interest
~ 60 % of cases

2.1. ppl who don't know, understand and/or care about the rules, claim something and disappear when asked to back it up + to respect the rules saying something negative about this place
~ 35 %

2.2. ppl who don't know, understand and/or care about the rules, claim something and provide samples and do ABX tests when asked to back it up + to respect the rules, resulting in them learning something and becoming valuable members of this community.
~ 5 %

From the 1st few posts of someone "2-ish" it's impossible to tell if (s)he's 2.1 or 2.2. type of person. This is an explanation for what you seem to regard as "low interest in problem samples" (not an excuse).


--------------------
Let's suppose that rain washes out a picnic. Who is feeling negative? The rain? Or YOU? What's causing the negative feeling? The rain or your reaction? - Anthony De Mello
Go to the top of the page
+Quote Post
Sonny
post Dec 8 2003, 12:07
Post #21





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



QUOTE (tigre @ Dec 8 2003, 01:40 AM)
Sonny, I tested with the (too long, see edit above) .ape sample you provided: lame 3.90.3 --aps, --aps -m s and couldn't hear anything obviously wrong. So please ...

- tell where exactly the problem is most obvious and what exactly it sounds like. Then upload an .ape sample < 30 seconds containing this position.

- perform an ABX test to prove that it's not just immagination

1. I didn't do ONE blind test, I've spent DAYS blind testing and made something like 10 tests, with up to 14 FILES and I always hear the distortion. Sometimes (for the 128kbps samples) I can hear it in less than 3 seconds. I begin lose the trace of that distortion when I change the command line, increasing encoding quality. In fact, using the '-V0', I wasn't able to hear anything distroted in high tunes (encoding in at nearly is at nearly 240 though)

2. tell you where is the distortion ? Did you actually listen to the 128kbps samples I've provided ? Surely not!! you wouldn't even ask. sad.gif

3. My imagination is great thank you biggrin.gif but, again I trust my ears. If I've passed 10 blind tests, it means I really HEAR there's something wrong!! When I write 'I'm not an axpert' it didn't mean : 'I'm deaf' !! rolleyes.gif
Anyway, to be sure that I was not a fool with vivid imagination, I did one more.

There was 'only' 6 files : the original wav, a 320kbps TS and four 128kbps samples. Three of them were encoded with 'alt preset cbr 128' command line. Two of them were encoded with Lame 3.93.1 one with JS the other with TS (adding '-m s' to the commadn line). The third, and that's new to me, with Lame 3.90.3. The fourth sample is a file downloaded from the internet, encoded with FhG in Joint Stereo.
I ran this test THREE TIMES. Results:
128kbps JS LAME 3.93.1 : found distorted on high tunes 3/3
128kbps TS LAME 3.93.1 : found clean on high tunes 3/3
128kbps JS LAME 3.90.3 : found distorted on high tunes 3/3
WAV original sample : found clean on high tunes 3/3
320kbps TS LAME 3.93.1 : found clean on high tunes 3/3
128kbps FhG JS : results are inconcluding

So... maybe the problem is coming from my computer, not from Lame, since some didn't noticed anything wrong using the 3.90.3 version... but I'm still not convinced : in fact, why do I hear distortion with Lame JS while there's nothing wrong in TS... unsure.gif
Go to the top of the page
+Quote Post
fairyliquidizer
post Dec 8 2003, 12:13
Post #22





Group: Members
Posts: 318
Joined: 11-November 03
Member No.: 9786



I just love the friendly supportive environment that is being nurtured here. Welcome to the Jungle, Sonny!


--------------------
http://www.glop.org/starforce/
Go to the top of the page
+Quote Post
Gabriel
post Dec 8 2003, 12:18
Post #23


LAME developer


Group: Developer
Posts: 2950
Joined: 1-October 01
From: Nanterre, France
Member No.: 138



Well, you are not using the winamp equalizer or any dsp plugin, are you?

If you want some feedback, please provide a short sample. 40s is way too long.
Go to the top of the page
+Quote Post
Sonny
post Dec 8 2003, 12:22
Post #24





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



QUOTE (Gabriel @ Dec 8 2003, 03:18 AM)
Well, you are not using the winamp equalizer or any dsp plugin, are you?

If you want some feedback, please provide a short sample. 40s is way too long.

I use Ultra Player, without equalizer. I also made tests with winamp 2.8 and 3.0, with and without equalizer....... same results.
Go to the top of the page
+Quote Post
Sonny
post Dec 8 2003, 12:22
Post #25





Group: Members
Posts: 20
Joined: 7-December 03
Member No.: 10256



QUOTE (fairyliquidizer @ Dec 8 2003, 03:13 AM)
I just love the friendly supportive environment that is being nurtured here.  Welcome to the Jungle, Sonny!

yeah i've seen...
Go to the top of the page
+Quote Post

3 Pages V   1 2 3 >
Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 2nd August 2014 - 04:38