IPB

Welcome Guest ( Log In | Register )

3 Pages V   1 2 3 >  
Reply to this topicStart new topic
Lame's future after --alt-presets
JohnV
post Nov 27 2002, 01:53
Post #1





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



QUOTE (Dibrom @ Nov 26 2002 - 10:03 AM)
Unfortunately I will not be making these presets. I am no longer working on LAME or with the LAME developers. Sorry. This is pretty much due to the endless conflict between the way LAME development currently works and my own ideology on the matter. I'd go more into detail, but I don't think there's much point in that.
I believe that JohnV may try to resume some of my work on presets, but you'll have to ask him.

Well, I don't think I want to "resume" Dibrom's work. Instead I'd want to revive the community testing for the future Lame versions, now that Dibrom has dropped supporting alt-presets.

Some of my ideas in no specific order:
1. I hope that 3.94 will make "non-forked" presets (using external presets) closer to a-p standard,extreme and insane quality levels possible.

2. I'd want the community to keep testing and giving recommendations for the future Lame presets if alt-preset is dropped. Lame 3.94 brings new quality improvements which need to be checked in the form of new switch testing.

3. This would definitely need the help of the "old gang": HansHeijden,Wombat,ff123,volcano etc.. I'd want that there could be somekind of "community recommendation" also in the future versions of Lame and more active testing. Testing should be done in normal ABX/DBT style and clear reports with original clips should be provided. Everything (every switch) should be justified. "wild style" switch suggestions would not be tolerated. Everything should be done in very systematic way.

4. Lame should ideally provide very high quality with simple switch sets. I'd hope that Takehiro would get some valuable hints, how to tweak the psymodel, when 3.90.2 APS/APX/API and 3.94 and later with best possible external switches are directly compared. I hope that in some point in the future no complicated forking is needed in order to gain APS/APX/API -quality and even beyond.

5. It should be checked especially for 3.94 and beyond if better external switches can be found than the switches in external preset A-Ps (non-APS/APX/API).

6. Imo it's important to keep supporting and testing Lame also in the future.

7. The recommended Lame version and presets should not change until it has been proven that something is better. However, now that Dibrom has dropped the support, it's important imo to keep up somekind of continuity for the future versions of Lame. In best case, this would lead to simple switch set which provides better results than APS/APX/API hopefully some beautiful day in the future. Until then, the official recommendation should remain the same. But at the same time for the sake of continuity, testing should be revived for 3.94 and beyond. Many people will always use the latest version, because they automatically expect quality improvements. Unless something is done, future Lame versions could infact go very much backwards in quality.

There.. I'd like to know what people think about this..


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
quellcore
post Nov 27 2002, 02:37
Post #2





Group: Members
Posts: 142
Joined: 14-June 02
Member No.: 2299



Sorry, don't have any useful things to add in this second, just want to mention that i'm kind of shocked about the news regarding Dibrom and his work on the presets in the future. At least i can understand his point, so many things regarding communication in the LAME projects didn't work as it should have.
Hopefully many people on this board will at least try to give some support with listeneing test, sounds like a nightmare to me that there could be a future version of LAME (let's say 4.0) with inferior quality. But as it seems to me right now it could happen easily! :'(


--------------------
Those who desire to give up freedom in order to gain security, will not have, nor do they deserve, either one! (Benjamin Franklin)
Go to the top of the page
+Quote Post
ff123
post Nov 27 2002, 02:38
Post #3


ABC/HR developer, ff123.net admin


Group: Developer (Donating)
Posts: 1396
Joined: 24-September 01
Member No.: 12



QUOTE (JohnV @ Nov 26 2002 - 04:53 PM)
2. I'd want the community to keep testing and giving recommendations for the future Lame presets if alt-preset is dropped. Lame 3.94 brings new quality improvements which need to be checked in the form of new switch testing.

3. This would definitely need the help of the "old gang": HansHeijden,Wombat,ff123,volcano etc.. I'd want that there could be somekind of "community recommendation" also in the future versions of Lame and more active testing. Testing should be done in normal ABX/DBT style and clear reports with original clips should be provided. Everything (every switch) should be justified. "wild style" switch suggestions would not be tolerated. Everything should be done in very systematic way.

...

There.. I'd like to know what people think about this..

I'd be willing to host samples as well as pitch in with listening opinions. There's a page around somewhere with the current list of problem samples for aps, isn't there? These and other samples used for regression testing would be useful as a reference.

It's going to take a pretty strong-willed individual to push through this type of agenda, though. I think you'd be a perfect guy to lead, JohnV.

ff123

This post has been edited by ff123: Nov 27 2002, 02:38
Go to the top of the page
+Quote Post
Negative Zero
post Nov 27 2002, 03:52
Post #4





Group: Members
Posts: 248
Joined: 16-March 02
From: Toronto, Ontario
Member No.: 1534



I'd just like to offer my sincere thanks to Dibrom for all of the time and hard work that he has put into making the --alt-presets a reality. Even if he has stopped development on them, the current --alt-presets with LAME 3.92 are good enough for me. It all sounds great on my SlimX! wink.gif

This post has been edited by Negative Zero: Nov 27 2002, 03:53


--------------------
http://www.welovetheiraqiinformationminister.com
"I triple-guarantee you this is the best web site ever!"
Go to the top of the page
+Quote Post
Dibrom
post Nov 27 2002, 04:10
Post #5


Founder


Group: Admin
Posts: 2958
Joined: 26-August 02
From: Nottingham, UK
Member No.: 1



Good luck on this, JohnV.

I have just a small bit of advice though: Before you get too far into planning for how you are going to approach this situation, I'd suggest having a very detailed discussion with the LAME developers about what exactly they are willing to do and what they aren't. It'd be wasteful to put a lot of effort into something only later to find out that the other people involved do not show the same motivation or share the same goals and interests as you. Also, don't assume that other people are going to follow your lead or go along with your plans just because they might be based on a good idea.

I think this is partially why my own approach didn't work out in the end.
Go to the top of the page
+Quote Post
floyd
post Nov 27 2002, 04:17
Post #6





Group: Members
Posts: 630
Joined: 18-June 02
Member No.: 2332



While its sad development on the presets is stopped (well, by Dibrom anyway), I personally never thought mp3 would sound as good as lame 3.9x ended up sounding - especially after horrible early experiences with Xing and Fhg - so three cheers for everyone who made it possible to get this far smile.gif And good luck to you, JohnV biggrin.gif
Go to the top of the page
+Quote Post
JohnV
post Nov 27 2002, 04:24
Post #7





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



Hmm.. I'd hope that people would not follow "my" lead, but there would be some core-group, which would give recommendations..

Here's a clear example why some of the current --alt-presets as such don't provide the best possible quality and what can be expected in the future from Lame.

Original awe32_20sec.flac

Current recommened Lame 3.90.2 --alt-preset cbr 128
awe32_3902-AP128cbr.mp3

Lame 3.94a2 --alt-preset cbr 128
awe32_394a2-AP128cbr.mp3

Lame 3.94a2 with simple -b128 -q0 --nspsytune (no tweaking, but gives a lot better quality than above)
awe32_394a2-128cbr.mp3

You can clearly notice, that the 3.94a2 --alt-preset cbr 128 is not optimal anymore at all and 3.90.2 also gives bad quality. I gained very much better results with very simple commandline in 3.94a2 which I didnt even try to tweak at all. This is why new tweaking is needed.

Btw. Mitiok does not provide the 3.94a2, but only 3.94a1 for some reason. 3.94a1 seems to lack important tweaks by Takehiro.
Lame3.94a2 is available here: http://static.hydrogenaudio.org/extra/LAME...-394-alpha2.zip

This example alone imo shows why testing and tweaking should continue.


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
ff123
post Nov 27 2002, 04:40
Post #8


ABC/HR developer, ff123.net admin


Group: Developer (Donating)
Posts: 1396
Joined: 24-September 01
Member No.: 12



QUOTE (JohnV @ Nov 26 2002 - 07:24 PM)
Hmm.. I'd hope that people would not follow "my" lead, but there would be some core-group, which would give recommendations..

IMO, it will take a strong leader with a good ear and at least a passing knowledge of how the code works to make sure that particular tweaks are tested out thoroughly. Otherwise, it will be hit and miss.

BTW, I think 3.90.2 sound clearly the worst on the sample clip you provided.

ff123
Go to the top of the page
+Quote Post
Dibrom
post Nov 27 2002, 04:50
Post #9


Founder


Group: Admin
Posts: 2958
Joined: 26-August 02
From: Nottingham, UK
Member No.: 1



QUOTE (JohnV @ Nov 26 2002 - 08:24 PM)
You can clearly notice, that the 3.94a2 --alt-preset cbr 128 is not optimal anymore at all and 3.90.2 also gives bad quality. I gained very much better results with very simple commandline in 3.94a2 which I didnt even try to tweak at all. This is why new tweaking is needed.

Of course. Let's not forget though that the cbr and abr modes of the --alt-presets never made use of code level tweaks. They are simply switch aliases (and also have not been tuned even close to that the VBR presets have been). It's no surprise that you can get better quality in some cases then by a different set of switches. VBR is a totally different story though, and I'm much more skeptical about getting better quality than the --alt-preset VBR modes via simple command line switches without very extensive help and motivation from one of the core LAME developers.

Edit: And btw, I haven't listened to the samples you provided.. so I'm not meaning to say I'm agreeing or disagreeing with you on that part. I only agree that higher quality is possible in some cases than the --alt-preset cbr and abr modes by using other command line options.

QUOTE
IMO, it will take a strong leader with a good ear and at least a passing knowledge of how the code works to make sure that particular tweaks are tested out thoroughly. Otherwise, it will be hit and miss.


I'd have to agree with this statement from ff123. I don't think a "consensus" approach is really going to work (it doesn't work in LAME development for this purpose at least). The reason that there were never --alt-presets in the past is because there was no leader to spearhead the effort and to try and get the quality tuning process consolidated and moving towards a specific goal. There are many people who are capable of working on something like what you are proposing here, but without a solid direction, like ff123 said, it'll be largely hit and miss. I don't think there are enough single highly motivated people (towards this goal) for it just "happen" coincidentally through everyone somehow sharing the same thoughts.

IMO, organization and structure are the key. The problem is getting people to agree with your ideas and to all work together towards the same goals.

This post has been edited by Dibrom: Nov 27 2002, 04:53
Go to the top of the page
+Quote Post
floyd
post Nov 27 2002, 04:52
Post #10





Group: Members
Posts: 630
Joined: 18-June 02
Member No.: 2332



those are some interesting samples, JohnV. I agree with your opinion that the untweaked 3.94a2 is superior.

3.90.2 - lots of distortion and flange on bass
3.94a2 --ap cbr 128 - less distortion, but chirps and watery sounds make this one the worst of the bunch.
3.94a2 untweaked - less distortion than both, small amount of chirping, seemingly only in the right channel..

Just for the heck of it I also tried 3.92 -ap 128 (abr) and MPC 1.14 --standard

3.92 - very close to 3.90.2 cbr surprisingly! In fact I believe slightly worse (more flange).
MPC 1.14 - transparent to my ears.. just notable for the fact that its 356 kbps! (probably someone pointed that out already)
Go to the top of the page
+Quote Post
JohnV
post Nov 27 2002, 05:18
Post #11





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



QUOTE (ff123 @ Nov 27 2002 - 05:40 AM)
BTW, I think 3.90.2 sound clearly the worst on the sample clip you provided.

Well, actually 3.94a2 --alt-preset cbr 128 sounds also very bad, because it has the same problem, which caused the 3.93 quality issue.. So the above awe32 example is not valid... Problem is broken quantization type (-q3 by default in A-P 128 in 3.94a2. Also notice that there were re-ordering of q-types. Lame 3.93 with -q2 uses the same q-type as 3.94a2 -q3, which are broken). So could be that --alt-preset cbr 128 would be the best switch when -q3 is fixed, but who knows..

Anyway, in Fatboy this is even more clear than in awe32:

Current recommened Lame 3.90.2 --alt-preset cbr 128
fatboy3902-AP128cbr.mp3

Lame 3.94a2 --alt-preset cbr 128 (quantization type -q3 broken, actually should ignore this sample)
fatboy394a2-AP128cbr.mp3

Lame 3.94a2 --alt-preset cbr 128 -q0
fatboy394a2-AP128cbr-q0.mp3

Here I provided only alt-preset cbr 128 samples, except the last example uses correctly working q-type (-q0). This shows how much better nspsytune pre-echo handling becomes in 3.94..

Heh.. at the same time I figured why Lame 3.93 sounded so bad.. its -q2 at least with cbr nspsytune is broken. Could be that Tak has already figured this out though. wink.gif


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
dev0
post Nov 27 2002, 06:31
Post #12





Group: Developer
Posts: 1679
Joined: 23-December 01
From: Germany
Member No.: 731



These are really sad news, since I really hoped for a preset for portables by Dibrom... sad.gif
Anyway, thanks for all the time and work you (and others) invested in the alt-presets, it's impressing hwat you did and how it succeeded.
The suggestions made by JohnV seem pretty sensible to me and I'll try to help whereever I can.
dev0
Go to the top of the page
+Quote Post
fewtch
post Nov 27 2002, 06:37
Post #13





Group: Members
Posts: 1460
Joined: 5-February 02
From: Seattle WA. USA
Member No.: 1261



Ditto... I think it's sad news too. Given the amazing prevalence of the MP3 format on portable devices of all sorts (and non-portable devices too) this isn't good news. Of course, the existing --alt-presets still work, but better tuning for lower bitrates is really needed in Lame, IMHO... not to mention, tuning for 48KHz. Hopefully someone will take the baton and run with it.

MP3 has a bad enough reputation among "audiophiles" as it is (especially the typical 128k CBR file), let's hope it doesn't get even worse now if the --alt-presets are broken in future Lame versions. It could even invalidate all of Dibrom's previous work and turn it into nothing but wasted time & energy -- by the time v4.0 of Lame is out, who will be using v3.90.2 except a small core contingent here on HydrogenAudio?

This post has been edited by fewtch: Nov 27 2002, 06:43


--------------------
Bring back dynamic range... www.loudnessrace.net
Go to the top of the page
+Quote Post
mithrandir
post Nov 27 2002, 07:31
Post #14





Group: Members
Posts: 669
Joined: 15-January 02
From: SE Pennsylvania
Member No.: 1032



QUOTE (JohnV @ Nov 26 2002 - 11:18 PM)
Here I provided only alt-preset cbr 128 samples, except the last example uses correctly working q-type (-q0). This shows how much better nspsytune pre-echo handling becomes in 3.94..

Heh.. at the same time I figured why Lame 3.93 sounded so bad.. its -q2 at least with cbr nspsytune is broken. Could be that Tak has already figured this out though. wink.gif

So if you use the VBR alt-presets with 3.94a2, you should use -q0 (overriding the default -q3)?
Go to the top of the page
+Quote Post
JohnV
post Nov 27 2002, 07:39
Post #15





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



QUOTE (mithrandir @ Nov 27 2002 - 08:31 AM)
So if you use the VBR alt-presets with 3.94a2, you should use -q0 (overriding the default -q3)?

I don't think APS should be used with 3.94a2 at all until somebody has verified it..


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
mithrandir
post Nov 27 2002, 07:43
Post #16





Group: Members
Posts: 669
Joined: 15-January 02
From: SE Pennsylvania
Member No.: 1032



QUOTE (JohnV @ Nov 27 2002 - 01:39 AM)
QUOTE (mithrandir @ Nov 27 2002 - 08:31 AM)
So if you use the VBR alt-presets with 3.94a2, you should use -q0 (overriding the default -q3)?

I don't think APS should be used with 3.94a2 at all until somebody has verified it..

I've been using it for some time - seems like it has the most advanced code of any compiled version. I don't remember hearing any nasties with APS using 3.94a2 but I haven't subjected it to any encoder-killing clips.

There is a major performance difference between -q0 and -q3, the latter is approximately 3x faster.
Go to the top of the page
+Quote Post
cookie
post Nov 27 2002, 08:52
Post #17





Group: Members
Posts: 120
Joined: 31-December 01
Member No.: 852



The question that comes up in my mind is:

Did did the fix that Gabriel made on 3.91 (so that 3.91.1 came out) go into 3.94a2 ?
3.93.1 right now has 'returned' to 3.92 quality, as it seems.


--------------------
sic transit gloria mundi...
Go to the top of the page
+Quote Post
JohnV
post Nov 27 2002, 09:26
Post #18





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



QUOTE (cookie @ Nov 27 2002 - 09:52 AM)
The question that comes up in my mind is:

Did did the fix that Gabriel made on 3.91 (so that 3.91.1 came out) go into 3.94a2 ?
3.93.1 right now has 'returned' to 3.92 quality, as it seems.

I have actually no idea what's the 3.93.1 version that Gaby provided for testing and what kind of modification he did to it, or is it just compiled from earlier source. It shows me 3.94a1..


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
JohnV
post Nov 27 2002, 09:42
Post #19





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



QUOTE (mithrandir @ Nov 27 2002 - 08:43 AM)
There is a major performance difference between -q0 and -q3, the latter is approximately 3x faster.

Well.. I'd guess that the -q3 issue does not necessary affect vbr.. I also tested 3.94a2 with some nspsytune lines and -q3 seemed quite fine. I haven't tested extensively so I'm not gonna guarantee anything...

btw. if you have used 3.94a2 with APS, what kind of bitrates it creates generally compared to 3.90.2 or 3.92?

This post has been edited by JohnV: Nov 27 2002, 09:51


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
Gabriel
post Nov 27 2002, 10:06
Post #20


LAME developer


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



I'd really like a kind of community effort for listening tests.
It's obvious that developpers can not have time to do every listening test (just speaking about time, not ability).
A community testing has a lot more time, because it's splitted between more people.

Listening test for every change are very time consuming, and I think that only a community is able to achieve this.

About lower bitrates presets: I think that I might do them, mainly because I'm interested in them. Also I have the hearing ability to do them, while I don't have this ability for high bitrates vbr presets.

Right now, the lower bitrate preset using vbr is medium. I received only few reports about it (but no negative reports).

The 3.93.1 version I provided in another thread for testing is a fixed 3.93, not an earlier version.
Go to the top of the page
+Quote Post
Wombat
post Nov 27 2002, 11:10
Post #21





Group: Members
Posts: 950
Joined: 7-October 01
Member No.: 235



Hello,

i have to admit i donīt have that much time at the moment for Lame.
But it gave me so much fun i will sure test further if i can.

JohnV as a coordinator makes sense and is very promising i believe.
On saturday i will go through aps testing with the newer versions.

cu

Wombat
Go to the top of the page
+Quote Post
JohnV
post Nov 27 2002, 11:45
Post #22





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



I tested how close Lame 3.90.2 --alt-preset standard I can get with quick external switch tweaking using Lame 3.94a2. (Mitiok does not host alpha2)

Anyway, this was the best I could do with external presets and 3.94a2 in short amount of time. It's not better than 3.90.2 APS in extreme pre-echo samples, but it's quite close, and it handles some of the APS problem clips better:
CODE
lame394a2 --nspsytune -V2 -mj --nsmsfix 1.1 -q3 -b112 --lowpass 19 --athtype 4 --ns-sfb21 2 -X0 -Z

Some explanation:
--nsmsfix 1.1: because of the serioustrouble ringing, cant be higher..
--athtype 4: same as base APS
--ns-sfb21 2: lower resolution (increase masking) for over 16khz freqs in order to combat bloat (because of the infamous no scf21 issue)
-X0: noise measurement type selected based on quick testing over -X1
-Z: switch for noiseshaping type1. Because of this some of the APS problem samples sound better.
-b112: lower min bitrate than in APS, because of ns-type 1 and -V2.
-q3: seems to work with vbr at least here.. although broken in cbr. Lots faster than -q0/-q1 and -q0 seems broken also.


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
mithrandir
post Nov 27 2002, 16:09
Post #23





Group: Members
Posts: 669
Joined: 15-January 02
From: SE Pennsylvania
Member No.: 1032



QUOTE (JohnV @ Nov 27 2002 - 03:42 AM)
QUOTE (mithrandir @ Nov 27 2002 - 08:43 AM)
There is a major performance difference between -q0 and -q3, the latter is approximately 3x faster.

Well.. I'd guess that the -q3 issue does not necessary affect vbr.. I also tested 3.94a2 with some nspsytune lines and -q3 seemed quite fine. I haven't tested extensively so I'm not gonna guarantee anything...

btw. if you have used 3.94a2 with APS, what kind of bitrates it creates generally compared to 3.90.2 or 3.92?

3.94a2 APS seems to have somewhat higher bitrates than 3.90.2 or 3.92 but it isn't enough to cause alarm. EncSpot reveals that 3.94a2 employs more short blocks than the previous versions, in some cases much more. I have several files where short blocks make up 15-20% of all blocks.
Go to the top of the page
+Quote Post
Nexx9
post Nov 27 2002, 17:45
Post #24





Group: Members
Posts: 36
Joined: 23-April 02
From: San Diego, California
Member No.: 1861



Is there any chance the situation will change to a more harmonious one, and is there any chance some developers can agree on an effort to get Dibrom back?

From the outside looking in it appears to me as if the entire preset effort is starting to unravel: Dibrom leaves because of a certain measure of disorganization, after which not even a Semi-organized effort is launched to keep him in the process, the implication being that either nobody gives a damn, or as a group they're so disorganized that nothing much matters except each person's 4.95638% of the project.

How wrong am I? Nex
Go to the top of the page
+Quote Post
Pio2001
post Nov 27 2002, 22:35
Post #25


Moderator


Group: Super Moderator
Posts: 3936
Joined: 29-September 01
Member No.: 73



I think that Dibrom deserves some good holidays, after all he has done. Let's not try to push him too hard for coming back for the time being.
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: 19th April 2014 - 06:50