Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Upcoming changes to --dm-presets - Part 2. (Read 31091 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Upcoming changes to --dm-presets - Part 2.

Alright.. some new developments.

I've created a new lower bitrate profile now which is very close to the --r3mix range.  Sometimes a few kbps higher, sometimes a few kbps lower.  In fact, on some clips it is up to around 20kbps lower!  The good news is that on pretty much all of the difficult test clips I have, it is outperforming --r3mix!  It isn't as high quality as --dm-preset standard but it's fairly close on some of these difficult clips.  Of course it's likely that this line can use some more tuning, and there may be a few quirks to work out that I haven't come across yet, but on the "killer" samples I normally use, it is performing really well.  That means it is much better on tonality issues, better on pre-echo, and better on impulses, often times all at lower bitrates.

Once I add it into CVS I highly encourage people to see if they can find some problems with it.. I've only been really testing it the last day or so myself.

I'm about to submit this new preset and along with this I will be changing the names of a few things.

--dm-preset will now be known as --alt-preset.

I'm doing this to remove my initials from the presets, and to promote the idea that they are truly alternate presets (as in an entire suite of presets now) and that they are not just some single idea that I am trying to promote.

--dm-preset standard will now become --alt-preset high

xtreme and insane will stay the same except xtreme will be changed to extreme.

This new mode will be called --alt-preset normal.

Before anyone asks, it does not use vbr-mtrh and it won't have a fast option for now either.  I may figure out a way to handle this later but I'm not sure yet.

Also, I've updated the --dm-preset fast standard profile to come closer in bitrate to --dm-preset standard.  --dm-preset fast xtreme is sometimes higher and sometimes lower than the non-fast counterpart, so I'm just leaving that alone.

I'm also working on maybe improving the impulse handling of --dm-preset standard... we'll see how that goes.

I'll post more details here shortly.

PS. Oh and btw.. I am aware that some people on other boards don't appreciate what I'm trying to do here and are resorting to referring to me as a communist of some sort or something to that effect.. *COUGH*cd-rw.org*COUGH* because I'm not wanting to explain the long command lines.. but that's ok.  The people who's opinions I really care about value this work and I think they understand the idea behind everything.

To the critics out there.. most of these tunings are made to work very specifically within the context of the other switches.. when you begin to unknowingly change things around then you lose the advantage of a highly tuned preset in the first place.  If that is how people are going to use all of this, then why am I bothering to tune this stuff so precisely in the first place?

Upcoming changes to --dm-presets - Part 2.

Reply #1 –
I'm very very impressed, Dibrom. There's just one small thing missing: what exactly is the commandline? Which modifications were made? I'm sure these will be answered in time, but somebody has to ask first anyway . . . might as well be me.

Thank you so much.

John

Upcoming changes to --dm-presets - Part 2.

Reply #2 –
I just want to clarify what Dibber is coughing about -> cd-rw.org/forum

Further developement? Excellent news. But as you see, it's nice to know in detail what actually is done.

Upcoming changes to --dm-presets - Part 2.

Reply #3 –
Once again awesome work Dibrom!!

Just a few questions....

(1) So did you make it in time for the new alt preset to be included in lame 3.90 stable?  My guess is yes, but i want confirmation...

(2) "It [alt normal] isn't as high quality as --dm-preset standard but it's fairly close on some of these difficult clips"

What about non-difficult clips?  Can you tell the difference between the new alt normal preset and the standard preset (now called alt high)?

Or to put it my way, If I encode Metallica with the new preset will it perform just as well as the old standard preset and still give me lesser filesize or are the differences noticible enough that you recommend metal-heads stay with the old preset?

(3) Does it include your new noise measuring -X9???

Upcoming changes to --dm-presets - Part 2.

Reply #4 –
My gratefulness goes to you, Dibrom!

With your kind assistance we now have a top-notch and easy to use preset system for LAME. No longer 320 characters long command line behemoths just to get decent quality, just a simple preset-commandline just like in MPC (and PsyTEL AAC) we're learned to love!

And I also support your idea of being not-exacly-very-open with the actual command lines. Neither MPC nor OGG offer us this much switches and tweaking possibilities and do I hear anyone complaining about that?
  If someone want's to know your commandlines so bad, let him look at parse.c just like everyone else .

One question though, will your tuned ABR modes work like in previous release, I mean like --alt-preset [bitrate] ?

Upcoming changes to --dm-presets - Part 2.

Reply #5 –
I'll answer multiple posts at once here..

Quote
Originally posted by Keynes
I'm very very impressed, Dibrom. There's just one small thing missing: what exactly is the commandline? Which modifications were made? I'm sure these will be answered in time, but somebody has to ask first anyway . . . might as well be me.


I'll be giving out the command line shortly, but it will be in a different manner than probably expected.

What I'm going to do from now on is use an alternate switch --dm-tune which will vary internal values depending on the V switch.  The reason is because the way it is now, when people see -V2 with dm-preset standard, they immediately think it must be lower quality because -V0 is not being used.  To make matters worse, the new line actually uses -V4 but is able to achieve the quality it does because of the other switches used alongside.. many people will not realize this though.

Because of this, I'm going to reconfigure the scale to make more sense.  It will be roughly equivalent to:

dm-xtreme = -V0 --dm-tune
dm-standard = -V1 --dm-tune
dm-normal = -V2 --dm-tune

All of this will of course be along with other stuff.. but it will allow me to give out the lines without all the confusing experimental options, and it will allow people to tweak the lines easier without doing something which may have a negative impact on quality.

-

Quote
Originally posted by RD

(1) So did you make it in time for the new alt preset to be included in lame 3.90 stable? My guess is yes, but i want confirmation...


Yes, and also the stable has been pushed back.  I think there is going to be a beta first now, and these changes will make it in time.

Quote
(2) "It [alt normal] isn't as high quality as --dm-preset standard but it's fairly close on some of these difficult clips" 

What about non-difficult clips? Can you tell the difference between the new alt normal preset and the standard preset (now called alt high)? 

Or to put it my way, If I encode Metallica with the new preset will it perform just as well as the old standard preset and still give me lesser filesize or are the differences noticible enough that you recommend metal-heads stay with the old preset?


This is the way I see it.. it is much easier to hear the differences on these difficult clips than it will be on easier clips.  This is kind of like what I was talking about with my original modifications.. with the fact that there was headroom there still but it was hard to remove that on the easy clips, without affecting the difficult clips.  I think this line is actually able to achieve that.  Of course, some of the difficult clips are degraded a little, but this new line actually sounds better than dm-standard on impulses!  dm-standard performs better on the tonality stuff.  Because of this, I'm going to work on the impulse handling with dm-standard a little so that there is not a weird drop in quality on some clips when going from normal to standard (high) to xtreme.  A lot of the work I have been doing lately is geared towards making the increase in quality among the profiles much more linear.. so this fits right a long with that.

Quote
(3) Does it include your new noise measuring -X9???


As of this very second, no.  But now that I have achieved this level of quality at lower bitrates, I can implement many of the ideas I originally had which were related to this.  I'm going to try and do that today before I submit this new preset.

-

Quote
Originally posted by niktheblak

One question though, will your tuned ABR modes work like in previous release, I mean like --alt-preset [bitrate] ?


Yep.. only the prefix to the presets will change.  dm becomes alt.

Oh, and btw.. thanks to everyone who offered words of support

Upcoming changes to --dm-presets - Part 2.

Reply #6 –
Quote
Originally posted by Dibrom

Oh, and btw.. thanks to everyone who offered words of support


I don't have much to say/ask but here is a post from me supporting you.

Upcoming changes to --dm-presets - Part 2.

Reply #7 –
Ok.. with the help of JohnV (thanks! ) I have been able to tune this preset even more.  I've implemented parts of this -X9 stuff I discussed before, and it led to further significant improvements in quality.  This preset is without a doubt superior on samples like fatboy, spahm, short, velvet, serioustrouble, 2nd_vent, death2, etc.. to name a few.

This X9 change has led to bitrate being increased slightly on difficult samples however, but overall the bitrate is low.. it seems as though often the bitrate is 10-15+kbps lower than --r3mix on most average clips.

I think once I've had more time, I can even fine tune the -X9 stuff to increase the bitrate even less than it does now (on the difficult samples), but still lead to the same improved quality.

Now what I need to do is just clean some of it up, encapsulate the experimental stuff within --dm-tune, and then I'll submit it to CVS.  This should happen before the end of the day, and I'll try to even include a version for download here, compiled with MSVC and ICL (intel).

Upcoming changes to --dm-presets - Part 2.

Reply #8 –
Great stuff. I'm amazed. That's all I can say at the moment

Since I used --r3mix for all my encodes until I changed to Vorbis, I am very interested in your "normal" preset.

CU

Dominic

Upcoming changes to --dm-presets - Part 2.

Reply #9 –
Hey Dibrom

thank you very very much for all your hard work!

CU

Upcoming changes to --dm-presets - Part 2.

Reply #10 –
Ok, for those interested in trying this out, the compile is up at:

http://static.hydrogenaudio.org/extra/lame_dm_rev5.zip

I adjusted the threshold of the X9 tweaks to where it doesn't quite increase the bitrate so much on difficult clips.  The quality seems to be mostly unaffected from this.

I'd very much appreciate feedback from anyone downloading this compile and testing it out.  If there are problems, I want to know about them so that I can try to fix them.

This compile includes the --dm-preset changes to --alt-preset

So to activate the new preset you just specify "--alt-preset normal"

Upcoming changes to --dm-presets - Part 2.

Reply #11 –
castanets normal bitrate: 176
castanets r3mix bitrate: 166

according to Winamp.  BTW, the bitrate calculation in Encspot 1.0 is broken because it reports 197 for the normal encode, which is wrong.

There is a big difference in pre-echo control on this sample using normal.  The difference between r3mix and normal is much larger than the difference between normal and original.

Trying to listen for other bad stuff besides transient smearing, I also tried wayitis.wav (20 sec version) and ringing.wav.  Of course, I never could hear anything bad with ringing.wav before, and I don't now either.  On wayitis (listening for noise pumping), I didn't notice anything bad.

Spectral views are very interesting, and don't look like what I've come to expect from mp3.  You can look at spectral views of normal and r3mix here:

http://ff123.net/export/ringing_normal.gif
http://ff123.net/export/ringing_r3mix.gif

When the high frequencies are quiet, normal is allowed to not encode even for frequencies below 16 kHz.  In such cases, I might want to listen for noise pumping and ringing.  But I'm not the best person for that job.  However, since Dibrom has tuned using SeriousTrouble.wav, I doubt ringing is a big issue.

ff123

Upcoming changes to --dm-presets - Part 2.

Reply #12 –
Yes, the behavior shown in these spectrograms is actually by design, however I was very surpised to actually see this as ff123 has shown it.  The idea behind this preset, and why it apparently works so well, is that it encodes areas which are normally deemed "more important" (high energy transients for example) with increased accuracy, and areas which are deemed not as important with less accuracy.

Looking at the spectrograms, you see there is more being encoded during the attacks, especially on the immediate decays, than --r3mix.  I've also listened to this clip and didn't notice the dropouts I'm used to in clips like serioustrouble and 2nd_vent.

On those particular clips, --r3mix allows more dropouts in the midrange frequencies while normal does not.  Again because of the difference in noise measuring.

Just thought I'd comment to clarify this behavior a little.

When I had originally talked about these noise measuring modifications, this is sort of what I had in mind.  Removing the headroom in certain areas of a file and concentrating more on the critical areas.  It appears that this is working.

Upcoming changes to --dm-presets - Part 2.

Reply #13 –
X9 really rocks!!!!

I don´t use your presets at the moment, but X 9 offers that many new possibilities - I love it!
Best of it, it doesn´t bloat unnecessaryly!

Wombat
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Upcoming changes to --dm-presets - Part 2.

Reply #14 –
Dibrom,

Quote
This compile includes the --dm-preset changes to --alt-preset 

So to activate the new preset you just specify "--alt-preset normal"


I downloaded the lame_dm.zip from the link you provided and used "--alt-preset normal", but it didn't work in razorlame.  It gave an error of an unrecognizable command and doesn't encode any mp3s.  Is the link you gave to a working version that you describe?

mp3

Upcoming changes to --dm-presets - Part 2.

Reply #15 –
Quote
This compile will give an error about a cygwin termcap on the command line, don't worry about it, it doesn't affect anything. Under win98 (because of this termcap stuff) the bitrate histogram may not be present either.  Razorlame will say there were errors encoding (because of the termcap message). None of this will be present in the normal compile, it is a cygwin only issue.


Check again, more than likely the mp3 is in the directory as it should be.

This compile does work, I've tried it, and apparently Wombat has also.

Upcoming changes to --dm-presets - Part 2.

Reply #16 –
Dibrom,

LAME_DM unzips into two files.  One is a lame.exe and the other is a .dll file.  I used Razorlame and the lame.exe encodes no music files.  The only file I get in the target directory is "LAME.EXE.stackdump".  What is that?

mp3

Upcoming changes to --dm-presets - Part 2.

Reply #17 –
Did you put the .dll in the same directory as LAME.exe?  You have to do this or it won't work correctly.  If that still doesn't work, I'm not sure what the problem would be.

What OS/Processor are you using?

Upcoming changes to --dm-presets - Part 2.

Reply #18 –
Dibrom,

This sounds very good indeed!

The --r3mix has had some critique for going too low with easy to encode material. You say that your new "normal" produces lower bitrates, but I presume that it is safer than --r3mix on material where --r3mix shoots very low?

Upcoming changes to --dm-presets - Part 2.

Reply #19 –
I have both files in the same directory.  I'm using Windows ME and an AMD K6 400, oc'd to 450, with Razorlame as a frontend which is using the lame.exe that came out of the zip.  Nothing should be wrong unless the proggy is buggy.

mp3

Upcoming changes to --dm-presets - Part 2.

Reply #20 –
Quote
Originally posted by cd-rw.org
Dibrom,

This sounds very good indeed!

The --r3mix has had some critique for going too low with easy to encode material. You say that your new "normal" produces lower bitrates, but I presume that it is safer than --r3mix on material where --r3mix shoots very low?


Yes, it should not go "too" low.

As a perfect example of this, death2 with --r3mix encodes near 112kbps or so, which is way too low considering the odd bass and extremely harsh transients in the sample.  "normal" encodes the file around 137kbps or so.

Also.. often times you may see 80kbps frames in areas with this new preset, but this (so far on the difficult samples I've been trying) doesn't really cause problems because of the stuff I just mentioned which keeps it from normally going too low in bitrate.

I think that probably near around 130 or 140kbps it will stop rolling off in bitrate so sharply unless whatever is being encoded truly does just have a ton of easy samples in it.  Stuff like movies should cause it to go lower, but during the difficult scenes (sound wise) it should scale right back up.

Upcoming changes to --dm-presets - Part 2.

Reply #21 –
Quote
Originally posted by mp3fan
I have both files in the same directory.  I'm using Windows ME and an AMD K6 400, oc'd to 450, with Razorlame as a frontend which is using the lame.exe that came out of the zip.


Try the program from the command line and see what is happening.

Quote
Nothing should be wrong unless the proggy is buggy.


I don't think so, none of my modifications are non-standard in the implementation within LAME and out of the 4 or 5 people so far that have tried it, you are the only one having problems.  I'm not sure anyone else was running Windows ME though, so the termcap issue with cygwin could be causing problems with that OS for some reason.  In that case, I'd say just wait for the MSVC version I'll try to have up shortly.. or wait for it to hit the mainline compiles.

Upcoming changes to --dm-presets - Part 2.

Reply #22 –
Here is the report Razorlame spits out:

Command: C:Program FilesLAMELAME DMlame.exe --alt-preset normal "D:WAV TEMPBiomechanical Theme - Biomechanical Theme.wav" "E:MP3 TempBiomechanical Theme - Biomechanical Theme.mp3"
      0 [main] lame 628837 handle_exceptions: Exception: STATUS_ILLEGAL_INSTRUCTION
  1149 [main] lame 628837 open_stackdumpfile: Dumping stack trace to LAME.EXE.stackdump
RazorLame encountered an unknown message from LAME while trying to encode "D:WAV TEMPBiomechanical Theme - Biomechanical Theme.wav"!

Encoded 0 files in 0:00:00
One file couldn't be encoded due to errors, probably wrong format.

mp3

Upcoming changes to --dm-presets - Part 2.

Reply #23 –
I use Windows SE, and had the problem with the non-displaying bitrate histogram; otherwise, the program worked ok.  But I don't normally use Razorlame either, preferring to work from the command line.

ff123

Upcoming changes to --dm-presets - Part 2.

Reply #24 –
I tried to run it in dos, but for some reason my dos prompt won't let it run.  Is there a special way to run it in dos?

mp3