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 31231 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

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

Reply #50 –
The files are not exploded in size that much.
I lowered the V value. Before i used --dm-preset standard -Z nsmsfix 1.17 and the line i use X9 with is like this but with V3 and lowpass 18.5 -b96.

This satisfies my needs.

The produced size is ok.

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 #51 –
Come on guys...

1. Dibrom is not forcing anyone to use presets.

He is working alone and cautiously to improve lame, and right now implementing some of his stuff as a new switch -X9 is not appropriate for the very good reasons he gave above, bitrate increases, dependency on specific switches, etc.

2. He is not an evil tyrant

As he has explained so well, and so many times, he has taken everyone's feedback seriously and tried to help tackle problems rather than just dismissing them... This is why --dm-preset standard is so well regarded as providing quality...

3. He has given so much effort and time unselfishly to help us and the lame project.

I for one have been amazed at how hard he works on lame, for no pay, and, all too often, with little gratitude from us...

I have often asked myself "Does this guy have a full time job?" because he works so much and so quickly I cannot fathom how he does it...

In fact, and I say this sincerely, his dedication and relentless drive to improve has lately inspired my work ethic.. and also made me feel like a slacker!  Few of us work as hard as him...

So lets be thankful...

Rome was not built in a day and to expect one man working largely alone to give perfect solutions and instant gratification on demand is unreasonable.

That being said I know many of you do appreciate Dibrom, wombat included.  We should all just show it more...

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

Reply #53 –
Well, I just tried out the new compile, using --alt-preset normal, versus the 11-11-01 compile with --dm-preset standard.

It worked fine, btw, on my 1.4Ghz P4 with Win2kPro.

I encoded Soft Cell - Memorabilia '91
with dm standard it averaged 212kbits, and in alt normal it averaged 180kbits.  I can't tell the difference, so I guess that is either great, or no help at all to you.

Thanks, Dibrom, for your great work.

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

Reply #54 –
Quote
Originally posted by wonderspark
I encoded Soft Cell - Memorabilia '91
with dm standard it averaged 212kbits, and in alt normal it averaged 180kbits.  I can't tell the difference, so I guess that is either great, or no help at all to you.


I'd say that was a good thing

Thanks for the feedback

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

Reply #55 –
Hmm,

In regards to "--alt-preset normal", did you implement your idea about short blocks being 320 kb/sec?  Or did you compromise on a lower possible minimum for short blocks?  I ask this because I'm not seeing 320kb/sec for short blocks on this setting, instead it seems to be selecting 224 kb/sec.  And there is a lower minimum bitrate as well.  The file I'm testing seems to be tapping many 80 kb/sec frames, however under the old dm-standard it tapped a lower average as well.  Just a few questions for ya. 

mp3

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

Reply #56 –
Quote
Originally posted by mp3fan
In regards to "--alt-preset normal", did you implement your idea about short blocks being 320 kb/sec?  Or did you compromise on a lower possible minimum for short blocks?  I ask this because I'm not seeing 320kb/sec for short blocks on this setting, instead it seems to be selecting 224 kb/sec.


Yes, I implemented parts of it.  I lowered from 320kbps to 224kbps after some listening tests.  The difference was very small but it lowered the bitrate a decent amount on difficult clips.  I compensated for this by allowing it to scale over 224kbps if it did need to and by using much more strict criteria for noise measuring.  For the high and extreme profile I'll be implementing similar ideas but with 256 and the 320kbps respectively.

Quote
And there is a lower minimum bitrate as well.  The file I'm testing seems to be tapping many 80 kb/sec frames, however under the old dm-standard it tapped a lower average as well.  Just a few questions for ya. 


Yes, the bitrate limit is 80kbps.  I decided upon this after testing quite a few samples along with JohnV and we found that it did not significantly impact quality once the short block stuff was in place.

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

Reply #57 –
So now I ran the same song a third time, using that link just posted (ICL+NASM version), and this time it went up to 185kbits average, but encoded much faster than the previous --alt-preset normal encode.  Yay!

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

Reply #58 –
Yeah, when using ICL the output is never completely identical to a cygwin compile, sometimes bitrates are slightly higher or slightly lower just due to different optimizations at the code level.

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

Reply #59 –
Dibrom,

Sorry for the interview style questions but, when you were talking about --dm-standard perhaps having room for more noise, is that where the lowering of -V2 to -V4 is coming into play? 

Also, I'm assuming you're using an nsmsfix setting of somewhere between 1.18-1.2, is that correct?  I changed the setting to nsmsfix 1.17 and found the sound to be a bit different, brighter.  I know that sounds goofy, but it just sounds different.  Could my different nsmsfix setting be trumping something you meant to leave in place?  I didn't compare either file to the original for accuracy, but they sound different when I change the command line.

mp3

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

Reply #60 –
Quote
Originally posted by mp3fan
Sorry for the interview style questions but, when you were talking about --dm-standard perhaps having room for more noise, is that where the lowering of -V2 to -V4 is coming into play?


Yes. 

Quote
Also, I'm assuming you're using an nsmsfix setting of somewhere between 1.18-1.2, is that correct?


Yes.

Quote
I changed the setting to nsmsfix 1.17 and found the sound to be a bit different, brighter.  I know that sounds goofy, but it just sounds different.


That's possible, though I'd need a clip and some abx results to verify.  On all except for some of the most difficult clips, usually transient heavy, there shouldn't be much of an obvious difference at all between nsmsfix values at those ranges.

Even then, when nsmsfix is set to high, a value of 1.17 isn't good enough.. you need 1.1 or lower.  One of the next things I'll be working on is some modifications to this threshold in certain situations.. but again, this is mostly only going to affect impulses.

Quote
Could my different nsmsfix setting be trumping something you meant to leave in place?  I didn't compare either file to the original for accuracy, but they sound different when I change the command line.


Well the point behind this line is efficiency.  That means that for most purposes, taking everything right to the threshold.  If you are changing values but one is not clearly better than the other, then I'm going to leave the value which offers a lower bitrate.  If you want more conservatism, you'd need to use a higher profile.  Thats the idea behind all of this.  I've already done a lot of work (with the help of JohnV) attempting to eliminate the most audible problems I can think of.. often times these are already going to be at a level most people will not be bothered with or even notice.. considering how many people have found --r3mix to be very good to them and that this switch is much higher quality.. usually very close to the old standard.

Also you have to remember that any modifications which affect bitrate as much as joint stereo thresholds, may not have so much of an effect on a particular clip, but may have huge effects on other clips.

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

Reply #61 –
I'll give my opinion about normal preset:
The "alt-preset normal" sounds generally very nice.
One issue which Dibrom is still trying to improve a bit at the moment is low volume accuracy without bitrate increase.

Code level tweaks can be done to control the bitrate even more. These are not very easy, but I have faith in Dibrom.

I talked with Naoki Shibata few days ago. He has been coding new tonality estimation implementation for Lame and for other (secret) project which uses same core tech for some time now. Also I really have to give credit to Dibrom, even Naoki said he ain't interested in figuring out presets because psychoacoustic model tweaking is so hard work. Of course Naoki and some other devs are the greatest, but to totally unleash the power of psychoac-model, a person like Dibrom is needed who tweaks at the code level.
Great work so far, and I know it will only become better.
Juha Laaksonheimo

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

Reply #62 –
Quote
Originally posted by Dibrom
The ICL+NASM version is up here:

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

It's much faster and should work fine on all systems.


as i see this version differs from cvs, so where i can get souce code (i preffer diff) for this binary?

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

Reply #63 –
Dibrom thanks again for your hardwork...

I don't know if your finished tweaking --al-preset normal... but one thing I did notice is that on some files the bass frequencies sound thinner than  using extreme (old standard + ath2 + X3).

For example I heard this in the first 30 seconds of Track 2 "It does not last" of Dark Tranquillity's album HAVEN.

The guitars sound thicker and more bass is there using extreme than with --alt-preset normal, which sounds thinner.

I don't know if this is worth fixing or if that is the point of using standard/high or extreme.

Last question, when you do eventually get to modify --alt-preset high will the average bitrate be lower than the old --dm-preset standard of lame 3.9a7?  I know on impulse music it will be better and probably higher in bitrate, but I wonder if metal music average bitrate will decrease a bit due to all your new tweaks...

Thanks.

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

Reply #64 –
Quote
Originally posted by RD
I don't know if your finished tweaking --al-preset normal... but one thing I did notice is that on some files the bass frequencies sound thinner than  using extreme (old standard + ath2 + X3).


Can you abx this?

The thing is, theres really no reason why bass frequencies should sound "thin" with what I'm doing here.

Quote
The guitars sound thicker and more bass is there using extreme than with --alt-preset normal, which sounds thinner.

I don't know if this is worth fixing or if that is the point of using standard/high or extreme.


Well I appreciate the feedback, but I'd ask you to try and verify this via ABX.  I can't really fix something if I don't know it's there

Seriously though, theres really nothing which is geared towards letting bass frequencies sound "thin" like that.  That isn't to say you aren't hearing a difference, but if you are it's likely from something else.  I'd ask if you can try to abx this, and if you can score higher than 12/16 (preferably 14/16), then it would be great if you could provide me with a sound clip.

Quote
Last question, when you do eventually get to modify --alt-preset high will the average bitrate be lower than the old --dm-preset standard of lame 3.9a7?  I know on impulse music it will be better and probably higher in bitrate, but I wonder if metal music average bitrate will decrease a bit due to all your new tweaks...


Yes, the average bitrate will probably be lower, especially on music like metal.

As it stands right now, this preset encodes metal with almost no bloating at all.. especially compared to other presets.  Routinely getting bitrates close to 40kbps lower on this music than --r3mix, are not uncommon.

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

Reply #65 –
Dibrom

I think i found a better sample that illustrates the point (track 8 same album)

I will do as you say with the abxing...

Unfortunatley my computer just died...the power suppy exploded and smels burnt after i pressed the power on switch... just happened 10 minutes ago... i am on my mom's computer...

So the abxing may take a few days...

Hopefully my other components are ok... otherwise i'm up sh*t's creek...

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

Reply #66 –
Dibrom,

I have been planning to write another LAME MP3 encoding article to: CD-RW.ORG, CDFREAKS.COM, CDR-ZONE.COM, CDRINFO.COM, but I have been pushing it since I wan't to wait until you get your developement into more stable phase (meaning - not changing weekly). Credits for the work will naturally go for you with links to this site, so your work will not go un-noticed in the scene.

Your "normal" seems to hit the spot perfectly as I hoped for in the LAME presets discussion. Well, no MTRH speed but we can live with that since the bitrates are VERY competitive. Great!

What I would really love to see is a really tight nutshell explanation (a line per implemented tweak) of the tweaks you have implemented, so that I could be able to explain for the readers why this stuff is good and how it differs from the old.

One thing you have made is the "short block tweak" that forces LAME to higher minimum bitrates (224->) on short blocks, if I got it right, but what else are the actual changes?

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

Reply #67 –
There is actually one other aspect that I am currently working on (which is why I haven't submitted to CVS yet), trying to implement some improvements.  I plan to explain all of this, but I'd rather wait until I'm sure its finished for now at least, otherwise I may say something works a certain way but in the finished product it isn't the same anymore.

It shouldn't be too much longer..

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

Reply #68 –
Hey Dibrom,

Here is one of the clips I'm talking about:

http://home.earthlink.net/~rad123b/test.zip

Test.zip contains two files:
----------------------------
(1) lpacv14.pac, (17.777 seconds)
archived with Lpac Archiver 1.4 (joint-stereo)
5-Extra High compression.

(2) alt-norm.mp3
the above clip encoded with your special
compile and --alt-preset normal

I think you'll hear what I am talking about,
namely that the bass frequencies/chunk of the
guitar sounds thinner in the mp3 than the original.

I abxed it 11/16 times, but my headphones are not
the best...

I interested to see what your ears come up with,
I am quite sure your ears are better than mine....

Of course if anyone else wants to test it, please download it and post your results...

PS I replaced the powersupply and all is well with the computer... whew!!!

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

Reply #69 –
Dibrom,

I trying out some orchestra score music.  So far, the bitrate choices seem very low during calm moments.  Like, under 128 constantly.  Is this what you expected to happen?  Or could it be that music with less volume is getting bit-starved?    Hope this is normal and expected.  I haven't ever found 96 kb/sec to sound good on much except maybe very simple keyboard-like tone stuff.  Just thought I'd run this by you.

mp3

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

Reply #70 –
RD,

I listened to that clip and didn't hear the problems you described.  I may listen again later, but there was no obvious "thinning" of frequencies to me which I think does make sense considering that this kind of thing shouldn't happen.

mp3fan,

On extremely low volume music, I am currently working on a modification to the athadjust code which will stablize bitrate a little more.  For the most part this music should be sounding OK, but there are some tweaks I'm implementing which will detect and encode these situations a little bit differently which should also raise the bitrate some.

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

Reply #71 –
Well,

I just listened to the encoded sample and then encoded the piece again with an older alpha 7 compile with --dm-preset standard.  On the "alt" preset, there was a flangy sound when the violas on the right channel broke out louder, but on the old "dm" standard there was no such flangy sound.  In fact I love how the "dm standard" sounded so much, I'm going to make my mp3s from that setting .  I really don't need to abx this one. 

Could there be time resolution issues with your setting considering the flange sounds?  It was kind of like putting it through a time-compression function or noise reduction function from soundforge or something.  Not highly noticable, but it was clearly there to me.  Otherwise, the quality was pretty darn good.

The pop music I've run through the new normal preset was pretty good and I didn't detect flange sounds, but then again, on louder music it wasn't picking so many 80-128 kb/sec frames.  Flange shouldn't be an issue if it was selecting those low bitrate frames correctly.

mp3

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

Reply #72 –
Quote
Originally posted by mp3fan
Could there be time resolution issues with your setting considering the flange sounds?  It was kind of like putting it through a time-compression function or noise reduction function from soundforge or something.  Not highly noticable, but it was clearly there to me.  Otherwise, the quality was pretty darn good.


No, this isn't what might have caused possible artifacts you may have heard (I'd still like abx scores when making statements about possible problems.. but anyway).

As stated, sometimes there may be possible problems on extremely low volume clip, especially those with hardly any transients.  This is due to the fact that currently the athadjust code does not compensate well enough in some of these situations.  This doesn't happen on all low volume clips either.  At any rate, I have a fix working but it needs to be fine tuned before I add it in to the preset.

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

Reply #73 –
RD: 11/16 is not very assuring result...
You could also try to abx against extreme if you think it's sounding ok.
Juha Laaksonheimo

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

Reply #74 –
I downloaded the sample and could not tell the difference.  I tried with both my Grados and my Beyerdynamic DT990s, the latter which have excellent bass response.

ff123