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: List of recommended LAME settings (Read 648587 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

List of recommended LAME settings

Reply #25
I've updated the list of settings to reflect the changes to the --dm-presets which have now been fully added to LAME.  Compiles with these changes should be out within a few hours to a day.

List of recommended LAME settings

Reply #26
Dibrom,

Great you implemented this preset system! Of course there are things i'm not happy about  so here it goes. Feel free to ignore, anyway I noted that additional switches behind the preset can be added!

The choise of lowpass: too high at low bitrates, too low at high bitrates.
There's a good reason to stay below 16 kHz at 128 k, because at this bitrate nspsytune's default -X1 doesn't encode > 16 kHz well, if at all. It tends to be smeared as well, though I don't recall on what samples. Encoding above 16 kHz is too ambitious period!
I suggest 80k/12.3kHz, 96k/15.3kHz, 112k/16kHz and 128k/16kHz.
For higher bitrates, lame's defaults are just fine:
160k/18.0kHz, 192k/19.5kHz, 224k/20kHz, 256k/21kHz, 320k/22kHz.

At 320k I noticed the use of both -X3 and -Z. Can you explain why the -Z as well? Unlike the combination with -X1, it does not make much difference to me.

I heard that you probably lowered the --ns-bass -8 value in the 128k preset. Clips like 41_30sec, spahm, youcantdothat etc rumble more than they already did. The -8 shall not be touched! :mad:

It would be nice to know where you transition to nssafejoint, -X3 etc.

Hans

List of recommended LAME settings

Reply #27
Quote
Originally posted by HansHeijden
There's a good reason to stay below 16 kHz at 128 k, because at this bitrate nspsytune's default -X1 doesn't encode > 16 kHz well, if at all. It tends to be smeared as well, though I don't recall on what samples. Encoding above 16 kHz is too ambitious period!
[/b]

Wombat disagrees with you on this case.  I think that a lowpass of 16khz is just a little bit too low.  Yes, you won't be able to encode much above that, but in cases when you can, it makes a difference to certain people.

Quote

At 320k I noticed the use of both -X3 and -Z. Can you explain why the -Z as well? Unlike the combination with -X1, it does not make much difference to me.
[/b]

The other day you were talking to me about using -Z even when it makes no audible difference usually but instead because on certain very artificial clips like test.wav or blips.wav it increases quality.  Since we are at 320kbps I decided to just use noise shaping 1 to sound better even in those cases.

Quote
I heard that you probably lowered the --ns-bass -8 value in the 128k preset. Clips like 41_30sec, spahm, youcantdothat etc rumble more than they already did. The -8 shall not be touched! :mad:


Way back when I did listening tests on certain clips I noted that a value of --ns-bass -8 changed the sounds of certain things like cymbals, even if it did improve the rumbling.  I'd rather make a more even trade off with a value of -6 then.

Quote
It would be nice to know where you transition to nssafejoint, -X3 etc.


At 192kbps I start using safejoint with an nsmsfix value of 1.7, at 224 I use 1.25, beyond that I just use the default of 1.  These nsmsfix values probably could use more tuning, I only did some quick and rough testing to decide upon them.  -X3 is turned on at 256kbps and up.

List of recommended LAME settings

Reply #28
Hans:

OK, I updated the lowpasses a bit:

  80kbps = 13500
  96kbps = 15300
112kbps = 16000
128kbps = 17500
160kbps = 18000
192kbps = 19500
224kbps = 20000
256kbps = 20500
320kbps = 21000

Also, a quite note about the --ns-bass stuff... if you are using Dmitry's compile, it only has the first set of modification I made included in it, which did NOT include ns-bass at all.  So if things are sounding really bad in that regards, that is why.  His compile came just a little bit too early.  Tomorrow everything should be fine though.  I'm going to go look for compiles which have all the latest changes (except these lowpass modifications of course).. I'll post a link here to the correct one to use today.

Edit2:  I take that back, smpman's compile isn't up to date yet either (no ns-bass tunings).  Looks like one more day..

Also, can you let me know which compile you were using Hans?  I'm just a little bit curious..

List of recommended LAME settings

Reply #29
Yes, I used Dmitry's compile, so that explains the huge 'rumble' deterioration I heard. The --ns-bass -6 you use instead of -8 should hardly be noticeable. Indeed --ns-bass takes away a few bits from the highs, but to my taste tackling the rumble is priority. In time, you may hear it more and more easily, because it's an unusual sounding artifact. While we hear HF artifacts immediately by now!

The lowpass table looks better now (except at 128 k  ), but I think there's a little error. At 256k you mention 20500 Hz, but that will give the same filter as 20000 Hz. The next one available is at 21000 Hz.

Forgot to ask, but I suppose you drop the --ns-bass at the same point where -X3 comes in?

Hans

List of recommended LAME settings

Reply #30
Quote
Originally posted by HansHeijden
Indeed --ns-bass takes away a few bits from the highs, but to my taste tackling the rumble is priority. In time, you may hear it more and more easily, because it's an unusual sounding artifact. While we hear HF artifacts immediately by now!


I can hear it, but yes it is a little unusual.  The more watery sounds of the highs tend to bother me more, but as you say, the --ns-bass -6 should work quite well while still allowing a bit more accuracy on the highs.

Quote
The lowpass table looks better now (except at 128 k  ), but I think there's a little error. At 256k you mention 20500 Hz, but that will give the same filter as 20000 Hz. The next one available is at 21000 Hz.


Ah.. forgot about that.  Hrmm, at 256kbps I'll probably drop down to 20000khz then.

Quote
Forgot to ask, but I suppose you drop the --ns-bass at the same point where -X3 comes in?


Actually 1 step before that.  I increase values from 80kbps up to 128kbps, then slowly back down till 224kbps where I stop using it completely, and at 256kbps I switch to X3.

List of recommended LAME settings

Reply #31
Updated November 6, 2001

Especially the dm settings should work since latest compile by Dmitry or SMPman of 6th of Nov. 2001 (20011106).

Hi ,

I added that sentence, so it should be clear with which version of lame the new settings work.

List of recommended LAME settings

Reply #32
Quote
Originally posted by Dibrom

  80kbps = 13500
  96kbps = 15300
112kbps = 16000
128kbps = 17500
160kbps = 18000


oh... imho lame can't give good quality with such lowpass, i'd like to have at least

  80kbps = 12000
  96kbps = 13500
112kbps = 15000
128kbps = 1600
160kbps = 17500

or even more strong lowpass.....

Quote
Also, a quite note about the --ns-bass stuff... if you are using Dmitry's compile, it only has the first set of modification I made included in it, 
<skip>
His compile came just a little bit too early.
<skip>
Edit2:  I take that back, smpman's compile isn't up to date yet either (no ns-bass tunings).  Looks like one more day..


once again, both smpman and i download sourcecode from ftp://cedric.vabo.cz/pub/linux/apps/lame/
so you always may check when Jan Rafaj made snapshot and what was made before snapshot.

Jerry Baker get source code directly from CVS, i did the same during lame3.80 and think that it's worse

List of recommended LAME settings

Reply #33
Added

List of recommended LAME settings

Reply #34
Hi,

I edited the resulting file sizes of VBR presets.

List of recommended LAME settings

Reply #35
I'd like to point out that on the r3mix.net version of this thread:

http://66.96.216.160/cgi-bin/YaBB.pl?board...4134851&start=0

The changes cd-rw.org made the "average" bitrates is flawed.  The reason ranges should not be given next to bitrates is because if you are to do that, you should do both extreme ends of the bitrate range.  I have tracks which go down to 128kbps with dm-standard, and I've also seen tracks which cause --r3mix to go over 260kbps.  Looking at the ranges on there now, it seems a bit skewed.  Maybe someone would want to comment on that over there, or change it back to averages only.

The original testing I did to determine rough average bitrates for the various presets were across a wide range of music (various forms of metal, various forms of electronic, experimental, ambient, spoken word, music with classical elements, etc) , with at least 20 cd.. this covered a large bitrate spectrum, I gather that the changes made to the presets listed there now are only from picking out a few tracks here and there and assuming particular behavior is common across most files.

List of recommended LAME settings

Reply #36
Dibrom,

Yes, the problem is that we can't really define the global average bitrate, in other words the global average for music.

The current figures are very rough estimates based on some average bitrates per album, not just one song. They can and will be finetuned, and now that you have optimized your settings further, some more testing needs to be done. I know that r3mix can go much higher and I know that standard can go much lower.

Still I think that for a popular music album there is a very good change that the bitrate will be inside the fork shown on the list.

EDIT: the list version here doesn't have the estimated bitrate forks, and the 170-200 comment on --r3mix is false since it easily goes much higher per album (see the r3mix forum version).

List of recommended LAME settings

Reply #37
I don't think it is a good idea to try and give ranges because it is impossible for that to be meaningful.  Those ranges are totally useless if someone only encodes classical, or only encodes metal.. 2 forms of music which are bound to cause drastically different bitrates for mp3.  A single average is even pushing it because that doesn't necessarily mean it will hold for the situations I just described.  The only averages I try to list are those had from encoding a very wide range of music and getting an estimate from that.  Single albums at a time should not be used, especially if they are from similar genres.

About the --r3mix range, I didn't create this list, though I'd rather even it would be removed.  The problem is I don't know know a more precise range for --r3mix because I haven't encoded a bunch of albums with it recently.  I'm planning on doing this "test" again though with all of the vbr presets I've created at which point I'll list more accurate figures.

Ranges are a very inaccurage thing with vbr (especially in MP3) and shouldn't be used to describe likely results, and I think having them even on the list on r3mix.net is misleading.

List of recommended LAME settings

Reply #38
What I think I'll do since I'm about ready to start reencoding again for average bitrates after some of these changes, is I'll keep the data for each album in the test "set".. and next to each preset I'll include a link to average bitrates across the albums encoded.  For each album in that set I can include a link to list the bitrates for the songs also.

That will give a much more accurate representation of the bitrates in various situations.

List of recommended LAME settings

Reply #39
The best way to represent a range would be to provide the mean and standard deviation, I think. People know that easier materials will yield lower bitrates, harder materials will yield higher bitrates. What they want to know is how much lower or higher the bitrate would be.

List of recommended LAME settings

Reply #40
Dibber: That's an idea but it will give only your average bitrate, so it is just as inaccurate globally.

And the bitrate estimates are nothing fundamental here. They just give a little guideline to the newbies and to how the presets differ. As more data is collected, they will be refined. And since you are on a fast developement pace, things need to be revised soon.

List of recommended LAME settings

Reply #41
Hi,

Let me start by saying that the new Lame settings are GREAT!
Dibrom, thanks a lot!

Secondly, user, Dibrom, are you planning to update the recommended Lame settings of this thread with the new --alt -preset options?

That would really help!
Jord.

List of recommended LAME settings

Reply #42
Quote
Originally posted by jord
Secondly, user, Dibrom, are you planning to update the recommended Lame settings of this thread with the new --alt -preset options?


Yep.. I'll be adding the --alt-presets as soon as they are in the mainline LAME which should be fairly soon.

List of recommended LAME settings

Reply #43
Great!

...and you'll add your compile to the mainline LAME as soon as it's final (with athadjust modifications completed), correct?

If so, I can't wait!
Thanks again for your work! :roflmao:

List of recommended LAME settings

Reply #44
--dm-preset <bitrate>

Simple question: In this tuned preset, <bitrate> has to be "standard"? Or can I use, for example, "--dm-preset 140"?

Thanks everybody for this thread, really useful, and should have a big red button in the front page!  Keep up the good work!

List of recommended LAME settings

Reply #45
Hi,
I just did.

See auldyin posts

auldyin

List of recommended LAME settings

Reply #46
Well, here is my offering.  I'd like the developers or Dibrom to consider this command line:

--nspsytune --nsmsfix 1.35 --abr 178 -b160 -m j -h --athtype 2 --lowpass 17.5

IMO, it uses only what it needs to achieve a great size/quality result.  It's not perfect, but IMO, the vast majority of users would be pleased with the quality from a "casual trade" perspective.  This command line has no real artifacts that I can hear, and it's definitely superior to a 192-CBR in my opinion because of it's flexibility.  When VBR bitrates get too bloated, I've been using this setting. 

Also, it uses a lowpass that conforms to most people's hearing ability.  Let's face it, the vast majority of people don't hear much beyond 16 KHz anyway.  I've found a 17.5 lowpass allows that little extra wisp of highs which is enough for even a pair of good ears. 

It all comes down to what you're willing to sacrifice when the bitrates go below about 192 kb/sec.  I put in what I could to maintain quality where it counted in that command line.  IMO, if you're willing to go below 192 kb/sec in bitrate, you're not really looking to preserve ulta-high frequencies anyway.  So, why not use a reasonable 17.5 KHz lowpass?  Makes sense to me.  Especially considering it's ABR, not VBR.

mp3

List of recommended LAME settings

Reply #47
why is every settin on this list still in the --dm format.

i thought it changed to --alt

and --alt-normal replaced the old --dm-standard.

what's the deal?

List of recommended LAME settings

Reply #48
Quote
Originally posted by SNYder
why is every settin on this list still in the --dm format.

i thought it changed to --alt

and --alt-normal replaced the old --dm-standard.

what's the deal?

As far as I know, the alt presets have not been submitted officially.
This would be why.

List of recommended LAME settings

Reply #49
Quote
Originally posted by Somebody

As far as I know, the alt presets have not been submitted officially.
This would be why.
So then how come --alt is already able to be used in LAME?

And do you ALMIGHT SOMEB0DY think it's fine for me to use --alt-normal over --dm-standard figuring he's done a hand full of stuff that I know are in --alt-normal but I'm not sure about being in --dm-standard? As in, are they the same exactly, and if not and --alt-normal is the latest work by him, is it ok to use that?

nglijzdgvbvg bidxbzvbgdkszbvgsabvkuhjvbls