(Thread split from: http://www.hydrogenaudio.org/forums/showth...ost&threadid=72 (http://www.hydrogenaudio.org/forums/showthread.php?goto=newpost&threadid=72))
Alright.. bumping this thread might be relevant due to some changes I've implemented in the --dm-preset switches in LAME(which I am still tweaking for a new revision btw). I will release a new compile as well as some cvs commits shortly.
Here are a few samples which have traditionally shown problems with the --dm-preset switches unless -Z was used. The -std is the standard --dm-preset, -std-Z is the standard --dm-preset with -Z, and -alt is the new "standard" dm-preset.
Short:
- Short with --dm-preset standard (http://www.animus-facticius.org/new_standard/short-std.mp3)
- Short with --dm-preset standard -Z (http://www.animus-facticius.org/new_standard/short-std-Z.mp3)
- Short with --dm-preset standard-alt (new mode) (http://www.animus-facticius.org/new_standard/short-alt.mp3)
Fatboy:
- Fatboy with --dm-preset standard (http://www.animus-facticius.org/new_standard/fatboy-std.mp3)
- Fatboy with --dm-preset standard -Z (http://www.animus-facticius.org/new_standard/fatboy-std-Z.mp3)
- Fatboy with --dm-preset standard-alt (new mode)
(See a few posts below for an updated sample)
Sample of "normal" music to show how similar the two modes are on non-difficult clips:
- 41_30sec with --dm-preset standard (http://www.animus-facticius.org/new_standard/41_30sec-std.mp3)
- 41_30sec with --dm-preset standard-alt (new mode) (http://www.animus-facticius.org/new_standard/41_30sec-alt.mp3)
The difference in quality on the more difficult samples is quite huge. Of course, these samples are still not perfect, and they probably will never be with MP3, but this is a big improvement in quality nevertheless.
The bitrate increase you see on the difficult samples is partially related to fixing the -Z issue, but it is also partially related to an attempt to improve pre-echo handling. I'm still working on the latter aspect though.
Another difficult sample which has been floating around recently:
- Ravebase with --dm-preset standard (http://www.animus-facticius.org/new_standard/ravebase-std.mp3)
- Ravebase with --dm-preset standard -Z (http://www.animus-facticius.org/new_standard/ravebase-std-Z.mp3)
- Ravebase with --dm-preset standard-alt (new mode) (http://www.animus-facticius.org/new_standard/ravebase-alt.mp3)
Hello all and dibrom,
Excuse when i hear that fatboy is even worse at least in the middlepart from 1.7 to 3.7 sec with the alternate compile contra std-Z and this when alternate even needs 38kbit more on average. :confused:
I can´t say the alternate mode does anything better with short than std-Z
But today it may not be my best day for do listening, i had nearly no sleep.
Wombat
I urge people not to test these yet..
Just heard the latest version of fatboy, and it's MUCH better!
Hi Wombat!
Since I've posted the last fatboy sample I've been tweaking lame some more and have come up with an even better sample, which you can get here:
http://www.animus-facticius.org/new_standa...boy-current.mp3 (http://www.animus-facticius.org/new_standard/fatboy-current.mp3)
Keep in mind, the idea isn't necessarily to come out better than with using -Z (though I think this version does) but instead to come to the quality level of -Z on these samples without having to use -Z on everything else. The problem is that -Z increases the bitrate by a large amount on almost all music, while this approach only causes LAME to scale up in bitrate on critical samples, but keep more normal and easy to code samples at nearly the same bitrate.
And as JohnV said, things are improving still. It's a work in progress folks, I'm just trying to give everyone a taste of what's in store
Last night, just before Dibrom posted his first message, he sent me fatboy, short, and 41_30sec to listen to, alt vs. std. I thought both fatboy and short improved noticeably, and 41_30sec did not degrade. Short had ugly background noises with std which alt improved.
Fhg VBR 100% still had the best fatboy, though.
ff123
Originally posted by ff123
Fhg VBR 100% still had the best fatboy, though.
Yeah, Fhg had the old version beat. The new one I just posted comes much closer to matching fastenc in quality. The beginning and ends of the file aren't as clean sounding as with fastenc, but fastenc has some artifacts in the middle that this lame version doesn't so its a much closer call than before I think.
@ff123
Yes alternate is ways better than standard, i talked about alternate wasn´t better than standard -Z!
Well, but how things can change! fatboy current gave me the first WOW effect since long - i still smile!!
Happy Wombat
__________________________________________________________
Edit: The short block behaviour is changed, much more of them in current
I just listened to fatboy-current, and it's very close to FhG VBR 100% now. Wombat says WOW with good reason -- this is quite an accomplishment.
ff123
Yes, yes, yes!
You have outdone yourself, Dibrom!
It sound even better than standard with -Z. How can this be? Are you fooling around with the vbr_mrth with -X0 in the latest alpha?
Keep it up. I'm sincerely impressed...
With the highest regards, stoff
Originally posted by stoff
Yes, yes, yes!
You have outdone yourself, Dibrom!
Thanks
It sound even better than standard with -Z. How can this be? Are you fooling around with the vbr_mrth with -X0 in the latest alpha?
Well it is a combination of a couple different things I'm doing. I'll explain it all once I've finished tuning everything and I release the compile. However, part of it is related to joint stereo, part of it is related to noise shaping 2, and part of it is related to short block handling. I'm not using vbr-mtrh or -X0 (the -X switch actually doesn't work with vbr-mtrh). Once I get this stuff under control, I do plan on looking at vbr-mtrh though.
Originally posted by stoff
Yes, yes, yes!
You have outdone yourself, Dibrom!
I second this.
I do notice a significant improvement with the new mode of fatboy over the standard sample. Sounds more defined.
in the short samples, the bass just sounds quite more deeper with your new mode.
Great job
Dibber,
Could you provide us with a compile, even a experimental one, for further testing?
Originally posted by cd-rw.org
Could you provide us with a compile, even a experimental one, for further testing?
Sorry but not yet. There are a few things I need to fine tune a little bit more before putting a compile out there. I don't think it should be too much longer though.
Originally posted by Dibrom
the -X switch actually doesn't work with vbr-mtrh
well, it works, but a little bit different compared to the old code. (no final mapping...)
Yeah I just noticed your latest cvs commits the other day which added this functionality. It should be interesting to see how it all works when finalized.
Hi Dibrom!
Just one question while we all eagerly await the "release": Will the tweak mainly affect bass response or should it also help short block handling on the higher tones? It would be awesome if the tweak also improved something like the hi-hat part of Roel's former test, e.g. less smeared?
Regards, stoff
Originally posted by stoff
Just one question while we all eagerly await the "release": Will the tweak mainly affect bass response or should it also help short block handling on the higher tones? It would be awesome if the tweak also improved something like the hi-hat part of Roel's former test, e.g. less smeared?
Well basically I am attempting to address both. The part about bass handling is for sure a given with these tweaks, the part about pre-echo on higher frequency transients isn't quite so easy though. Already the --dm-preset's handle pre-echo better than alternatives (say --r3mix for example) but I'd like to get it even better if possible. We will just have to see.. I have tweaked LAME to where it produces better results on death2 (very harsh pre and post-echo test case) and castanets but I am not quite sure if this tweak can be applied appropriately in other situations yet.
Basically, I'm just trying a bunch of different things at the moment. Information you may see elsewhere as to what exactly I am doing is most likely out of date. The only thing to say is just "wait and see" really
Dibrom,
Thanks again for all your hardwork with these compiles... you have acomplished more in the last 2 months than r3mix accomplished in a year!
Now my next question is probably mundane but here it is anyway:
Where on this site are you going to post your special compile for download?
On this:
http://www.hydrogenaudio.org/download.php (http://www.hydrogenaudio.org/download.php)?
(which requires an account)
On animusfacticius.org? (probably spelled it incorrectly and I know latin too--artificial soul/mind?) Is this site still operational or did i just mispell it?
Or where?
Thanks,
eagerly awaiting your compile,
RD
Originally posted by RD
Thanks again for all your hardwork with these compiles... you have acomplished more in the last 2 months than r3mix accomplished in a year!
Thanks
Now my next question is probably mundane but here it is anyway:
Where on this site are you going to post your special compile for download?
On this:
http://www.hydrogenaudio.org/download.php (http://www.hydrogenaudio.org/download.php)?
(which requires an account)
On animusfacticius.org? (probably spelled it incorrectly and I know latin too--artificial soul/mind?) Is this site still operational or did i just mispell it?
Or where?
I'll probably host the file on http://www.animus-facticius.org (http://www.animus-facticius.org) (note the -, and you basically got the meaning right.. artificial consciousness/intelligence is a bit more in the vein of what I was going for but close enough.. it is actually the name of another project which I haven't really opened up to the public (unrelated to audio encoding) at this point.. so I just host stuff from there instead)... as well as provide a link in the downloads section. I don't imagine this will stay up too long though since most of the changes will probably be incorporated back into LAME.
...ah... its all becoming clear now...
you are a musician as well...the green image in your posts is the "artificial conciousness" and that animus-facticius.org is the website that was planned for your music release...
Am i right?
Man what a prodigy: musician/programmer/audiophile/golden-ear/beer-expert/ where does it end dibrom?!?!
Well i am a musician too--would like to hear your stuff...
RD
Originally posted by RD
Am i right?
Yeah.. pretty much
Well i am a musician too--would like to hear your stuff...
Perhaps I can send something sometime. Unfortunately in the last few months I haven't really been in much of a situation to work on the project to the extent I would have liked, but hopefully in the coming months that will change.
Alright.. I finally got one of the bigger tweaks I was working on for all of this in place now Basically, it is an entirely new frequency based noise measuring criterion (-X9) which should provide quite an increase in quality on impulse samples, and I'm hoping on just transients in general. It is different from the other modes in that depending on the current scalefactor band, it uses different levels of noise measuring (maximum, total+maximum, average, etc).
I'm polishing it all up now and then I'm going to add the rest of the tweaks that I've been working on.
It looks like there is probably going to be a test version of my new compile available sometime today... so stay tuned
Nervously shaking and cleaning ears!
Originally posted by Wombat
Nervously shaking and cleaning ears!
Same!
And so do I!
Argh.. well it looks like there is going to need to be a little bit more tuning before I release this after all
Everything is working great, but a few samples are causing rather large bitrate increases (not related to transients or pulses or any of that), so I'm trying to get that back under control first. I think a release is imminent, but not quite yet. I'll post again once it is under control (which I hope should be "very soon now" )
Hi Dibrom...
Any progress? I'm sorry for being impatient here, but the tweaks just sounds so darn interesting! :listening
Regards, stoff
I know how you feel Stoff, because I just got the Rockford Fosgate RFX9000 car mp3 player and it rocks!!...highly recommended...so I cannot wait to use Dibrom's special compile to **finally** store my 500 cds onto mp3-cds.
So far --dm-preset-standard -Z sounds awesome in the car, but I've waited this long, so I'm waiting for the "special compile".
Btw, thanks again for all your hardwork Dibrom...
RD
Originally posted by stoff
Any progress? I'm sorry for being impatient here, but the tweaks just sounds so darn interesting! :listening
Yes, well sorta I've been trying these modifications a couple different ways so far and each time, I find a way to where I think it will work just a little bit better. Right now I think I've come to what might be the best implementation I've thought up yet, but I'm still working on putting it all together. What I've decided is that if I can't get some of this working in a short period of time, I'll probably just put out some sort of interim release to keep people occupied in the meantime
I'm going to try hard to finish this stuff up this weekend..
Hmmm... Taking a look at your new presets Dibrom, I'm wondering what the difference between --dm-preset insane & --dm-preset cbr 320 is? I tested these presets with a sample (the intro from the album "Schizophrenia" by Sepultura), and this is what EncSpot 1.0 came up with upon looking at the details of the MP3s:
Sepultura - 01 - Intro (CBR 320).mp3
------------------------------------
Bitrates:
----------------------------------------------------
320 |||||||||||||||||||||||||||||||||||||||| 100.0%
----------------------------------------------------
Type : mpeg 1 layer III
Bitrate : 320
Mode : joint stereo
Frequency : 44100 Hz
Frames : 1228
Length : 00:00:32
Max Reservoir : 396
Emphasis : none
Scalefac : 17%
Complete : yes
Encoder : Lame 3.90
--[ EncSpot 1.0 ]--[ http://www.guerillasoft.com (http://www.guerillasoft.com) ]--
Sepultura - 01 - Intro (Insane).mp3
-----------------------------------
Bitrates:
----------------------------------------------------
320 |||||||||||||||||||||||||||||||||||||||| 100.0%
----------------------------------------------------
Type : mpeg 1 layer III
Bitrate : 320
Mode : joint stereo
Frequency : 44100 Hz
Frames : 1228
Length : 00:00:32
Max Reservoir : 396
Emphasis : none
Scalefac : 36%
Complete : yes
Encoder : Lame 3.90
--[ EncSpot 1.0 ]--[ http://www.guerillasoft.com (http://www.guerillasoft.com) ]--
Is the only difference between these two presets a higher scale factor?
I know I can find out what switches these presets actually consist of at the CVS of LAME <http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/lame/ (http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/lame/) >, but I can't figure out the system there. I guess I'm stupid or something. =)
Originally posted by kjempen
Is the only difference between these two presets a higher scale factor?
This would appear to be the case. I suspect one is using noise shaping type 1 and the other type 2. I'll look into this. They should be identical otherwise..