IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
Last Version of Lame only with gpsycho
Jerloch
post Dec 14 2006, 16:36
Post #1





Group: Members
Posts: 12
Joined: 7-September 05
Member No.: 24367



What was the last verision of lame before nspsytune was introduced ? why whas gpsycho overtaken by nspsytune ?

Is it just me or does gpsycho sound better ?

How can i succefully use gpsycho in lame 3.90.3 ?

What does the psycho acoustic model do anyway ? save bitrate ? or does it have a real impact on the sound
itself ?
Go to the top of the page
+Quote Post
halb27
post Dec 14 2006, 16:51
Post #2





Group: Members
Posts: 2424
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



You can use 3.90.3 with GPSYCHO and ABR for instance by using --abr 270 -h.
I used this setting a lot some time ago, and quality was great to me even with the severe problem samples I tried for a long time.
However: you don't get joint stereo with this setting, and this was the reason why I gave up this usage.
(There was a pre-echo sample with serious flaws).
I didn't dare to force joint-stereo with GPSYCHO as safe joint stereo was introduced with nspsytune AFAIK.
Moreover at ~250 kbps there's not much difference qualitywise between GPSYCHO and nspsytune.
Last not least 3.90 development with it's signficant testing and tuning was done for nspsytune.

So taking it all together I think it's wiser to actually use nspsytune though I would have preferred GPSYCHO development to have continued (the problems current 3.98 is about to solve were not there when using 3.90.3 GPSYCHO abr or cbr) - but it's not like that and we have to face it.

This post has been edited by halb27: Dec 14 2006, 16:51


--------------------
lame3100m -V1 --insane-factor 0.75
Go to the top of the page
+Quote Post
gameplaya15143
post Dec 16 2006, 17:57
Post #3





Group: Members
Posts: 484
Joined: 8-January 06
From: Earth
Member No.: 26978



nspsytune was an OPTION from 3.93.1 and earlier. By default 3.93.1 and earlier use gpsycho, to use nspsytune you need to either use an --alt-preset or add --nspsytune to the commandline. Lame 3.94 and later all use nspsytune(2?) by default, and there is no option to use gpsycho instead sad.gif (I wish there was).

Does gpsytune sound better? .. it depends on the track you are encoding. For low bitrates I greatly prefer nspsytune. At high bitrates, since these older encoders sill have some bad problem samples with nspsytune (and it's slower than gpsycho), I generally prefer to use gpsycho. You'll have to do some testing and decide for yourself which is better for you.

If you are at all worried about problem samples, I would suggest using the latest version of lame, as most of the problems have been, or are being, fixed.

If you want to know what a psycho-acoustic model does.. use --athonly (disables the psymodel) and listen wink.gif


--------------------
Vorbis-q0-lowpass99
lame3.93.1-q5-V9-k-nspsytune
Go to the top of the page
+Quote Post
robert
post Dec 16 2006, 19:20
Post #4


LAME developer


Group: Developer
Posts: 788
Joined: 22-September 01
Member No.: 5



QUOTE (gameplaya15143 @ Dec 16 2006, 17:57) *
Lame 3.94 and later all use nspsytune(2?) by default, and there is no option to use gpsycho instead sad.gif (I wish there was).
try LAME 398a8 with --psymodel 1, this switch is available with alpha builds.

QUOTE
If you want to know what a psycho-acoustic model does.. use --athonly (disables the psymodel) and listen wink.gif
if you want to disable the psymodel, you would have to do: lame -q9 good.wav bad.mp3
Go to the top of the page
+Quote Post
memomai
post Dec 16 2006, 19:49
Post #5





Group: Members
Posts: 264
Joined: 13-February 05
From: Germany, Kempten
Member No.: 19808



QUOTE
try LAME 398a8 with --psymodel 1, this switch is available with alpha builds.


Do I something wrong or why doesn't work the encoding process with 398a8 at me using -b 192 --psymodel 1?

EDIT: oh sorry I must have done something wrong wink.gif 398a8 seems not to work with old razor lame XD

now EncSpot shows me gpsycho instead of nspsytune smile.gif it's time to make some tests smile.gif

This post has been edited by memomai: Dec 16 2006, 19:54


--------------------
FB2K,APE&LAME
Go to the top of the page
+Quote Post
memomai
post Dec 16 2006, 20:29
Post #6





Group: Members
Posts: 264
Joined: 13-February 05
From: Germany, Kempten
Member No.: 19808



QUOTE
You can use 3.90.3 with GPSYCHO and ABR for instance by using --abr 270 -h.
I used this setting a lot some time ago, and quality was great to me even with the severe problem samples I tried for a long time.
However: you don't get joint stereo with this setting, and this was the reason why I gave up this usage.
(There was a pre-echo sample with serious flaws).
I didn't dare to force joint-stereo with GPSYCHO as safe joint stereo was introduced with nspsytune AFAIK.
Moreover at ~250 kbps there's not much difference qualitywise between GPSYCHO and nspsytune.
Last not least 3.90 development with it's signficant testing and tuning was done for nspsytune.

So taking it all together I think it's wiser to actually use nspsytune though I would have preferred GPSYCHO development to have continued (the problems current 3.98 is about to solve were not there when using 3.90.3 GPSYCHO abr or cbr) - but it's not like that and we have to face it.


you're right halb. gpsycho isn't developed for the new versions of lame. well, I think nspsytune does a great job at all, although gpsycho would have prevented some problem samples...

If you want to use gpsycho use lame 3.90 till 3.93.1 without tuned options, like -b XXX, --abr XXX and so on (no --presets / --alt-presets)

QUOTE
What does the psycho acoustic model do anyway ? save bitrate ? or does it have a real impact on the sound
itself ?


psymodels are there to "trick" the human ears. Using such low bitrates, the psymodel has to tune the sound to make the listener feel he listens to an original cd.


--------------------
FB2K,APE&LAME
Go to the top of the page
+Quote Post
Jerloch
post Dec 16 2006, 22:14
Post #7





Group: Members
Posts: 12
Joined: 7-September 05
Member No.: 24367



Thank you for your replies.

It seems that my hearing is "tuned" to gpsycho , To me it gives the right sound.
Maybe it has something to do with the kind of music im listening to or maybe its the shape of my ears ?

only thing im worried about is that Safe joint stereo is only usable with nspsytune , am i right ?
i have used the --nssafejoint and the lame header says safejoint stereo = yes but when tested with encspot
the m/s - Stereo frames are no different at all when using gpsycho the percentage of the m/s - stereo stays exactly the same when using joint stereo with or without --nssafejoint

On the other hand i have done some listening test with music pieces that have very big stereo separation in certain parts of the songs and gpsycho joint stereo seems to work ok with that.

i have listened to the mp3,s and then compared with the cd ripped wav file and turning the balance all the way to the left and right and it works ok.. i cannot detect any channel "leeking" between the left and right channel. that was on music passages that have very distinct sounds appearing only to the left or right.

This post has been edited by Jerloch: Dec 16 2006, 23:09
Go to the top of the page
+Quote Post
Cygnus X1
post Dec 16 2006, 23:40
Post #8





Group: Members (Donating)
Posts: 676
Joined: 5-June 02
From: New York
Member No.: 2224



QUOTE (Jerloch @ Dec 16 2006, 16:14) *
Thank you for your replies.

It seems that my hearing is "tuned" to gpsycho , To me it gives the right sound.
Maybe it has something to do with the kind of music im listening to or maybe its the shape of my ears ?


What exactly about --nspsytune isn't "right"? Pleae provide some samples and/or blind testing results of instances where gpsycho is better than nspsytune - I'm sure the LAME developers would be interested.
Go to the top of the page
+Quote Post
halb27
post Dec 17 2006, 00:01
Post #9





Group: Members
Posts: 2424
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



QUOTE (Cygnus X1 @ Dec 17 2006, 00:40) *
What exactly about --nspsytune isn't "right"? Pleae provide some samples and/or blind testing results of instances where gpsycho is better than nspsytune - I'm sure the LAME developers would be interested.

herding_calls and trumpet for instance at a bitrate of ~190 kbps. GPSYCHO provides the better quality. This is even true at ~250 kbps but the difference is negligible there.
Make sure to use ABR - VBR like aps is simply horrible.


--------------------
lame3100m -V1 --insane-factor 0.75
Go to the top of the page
+Quote Post
halb27
post Dec 17 2006, 00:17
Post #10





Group: Members
Posts: 2424
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



QUOTE (Jerloch @ Dec 16 2006, 23:14) *
... i cannot detect any channel "leeking" between the left and right channel. that was on music passages that have very distinct sounds appearing only to the left or right.

I think it will only occur on rare occasion that joint stereo can provide a problem. That's why joint stereo is the way to go. The improved degree of security provided by 'safe joint stereo' however is restricted to nspsytune.

I personally see it this way:

If you want to use a bitrate of ~250 kbps or more: use for instance --alt-preset 270 which means you use nspsytune and safe joint stereo. At such a bitrate I guess you won't find a sample where GPSYCHO behaves significantly better.

If you want to use way below 250 kbps, say 190...220 kbps then GPSYCHO is preferable with respect to those samples I have in mind. So I personally feel with your preference for GPSYCHO though I know the problems I have in mind are not the universe of music. Your forcing of joint stereo is to be preferred IMO against plain stereo mode.


--------------------
lame3100m -V1 --insane-factor 0.75
Go to the top of the page
+Quote Post
Jerloch
post Dec 17 2006, 00:44
Post #11





Group: Members
Posts: 12
Joined: 7-September 05
Member No.: 24367



So setting --nssafejoint with gpsycho only sets the flag in the lame header to "safejoint stereo = yes" and nothing else ?
Go to the top of the page
+Quote Post
Kirby54925
post Dec 17 2006, 01:15
Post #12





Group: Members
Posts: 64
Joined: 22-July 05
Member No.: 23461



As Cygnus said, please provide some test results before proceeding any further. As of now, I believe you're breaking TOS #8 by claiming that your ears are "better-tuned" for gpsycho.
Go to the top of the page
+Quote Post
halb27
post Dec 17 2006, 10:14
Post #13





Group: Members
Posts: 2424
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



QUOTE (Jerloch @ Dec 17 2006, 01:44) *
So setting --nssafejoint with gpsycho only sets the flag in the lame header to "safejoint stereo = yes" and nothing else ?

Yes.
I just confirmed this encoding 3.90.3 --abr 270 -h -m j with and without --nssafejoint.
According to foobar the encodings are bit-identical as was expected.

As for joint stereo it's like this in theory:

The l/r and m/s representation of the stereo signal are equivalent, and such is the mixture of the l/r amd m/s representation aka joint stereo.
With lossless encodings there can be absolutely no problem using joint stereo.
With lossy encodings it's different, and it can affect more than just stereo separation:
a) the switching point from l/r to m/s representation and vice versa can cause audible problems.
b) in m/s representation the usual psy model applies to the m representation. This is not quite so for the s representation which is a difference signal. So another psy model is needed. Any psy model can have weaknesses, and with two psy models we get two sources for possible weaknesses.

So much for the theory.

In practice there's not much to be worried about. You can find a lot of topics here on HA related to joint stereo problems. But nearly all of them are theoretical reasoning. Compared to that practical problems are next to nothing.

So your approach using joint stereo with 3.90.3 GPSYCHO may be well allright also without --nssafejoint.

This post has been edited by halb27: Dec 17 2006, 18:05


--------------------
lame3100m -V1 --insane-factor 0.75
Go to the top of the page
+Quote Post
halb27
post Dec 17 2006, 18:56
Post #14





Group: Members
Posts: 2424
Joined: 9-October 05
From: Dormagen, Germany
Member No.: 25015



QUOTE (Kirby54925 @ Dec 17 2006, 02:15) *
As Cygnus said, please provide some test results before proceeding any further. As of now, I believe you're breaking TOS #8 by claiming that your ears are "better-tuned" for gpsycho.

If you like to I can provide an abx result for herding_calls, but as you're sceptical you better listen for yourself.
3.90.3 abr 190 gpsycho isn't transparent but better than abr 190 nspsytune.
At a bitrate like 250 kbps the difference is still there but neglegible.
You can use cbr 192 instead of abr 190 if you like to.
VBR is horrible no matter whether gpsycho or nspsytune. Don't use aps.
3.97 makes things worse, so far 3.98 as well.

Anyway this one sample doesn't prove anything. It just backsup a personal preference for gpsycho somebody might have. There's no such thing as a proof or even hard evidence to prefer gpsycho or nspsytune one over the other, especially at high bitrate. Common preference is for nspsytune, and there are good arguments for this preference. But if somebody prefers gpsycho there's nothing wrong with it even if he only feels that gpsycho suits him better (as long he doesn't start convincing other people based on his feelings).

Some time ago I studied what was going on with the developers' primary support going from gpsycho to nspsytunes. There was no clear advantage of nspsytune then, it was just that Dibrom who seems to have been the most energetic developer then was in favor of nspsytune, probably because ringing problems were in focus for which nspsytunes promised to bring advantages. JohnV was in favor of gpsycho.
But as Dibrom was the driving force of Lame development development continued using nspsytune. This is how things go, it's allright.

This post has been edited by halb27: Dec 17 2006, 18:59


--------------------
lame3100m -V1 --insane-factor 0.75
Go to the top of the page
+Quote Post

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: 25th July 2014 - 22:17