Skip to main content
Topic: Lame 3.98 beta 6 (Read 138568 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Lame 3.98 beta 6

Reply #125
You can have normal mono/stereo wave files with WAVE_FORMAT_EXTENSIBLE format definition. Previously LAME did reject them or interpreted them as raw PCM files. Now LAME recognizes and may read them.

OK - thanks a lot .

Arite.

Lame 3.98 beta 6

Reply #126
Quote
Wrong about Vorbis. It is absolutely necessary to keep Vorbis I as stable and the codebase as constant as possible.


Vorbis I is not even the recommended Vorbis encoder in HA. All spotlight is directing to the only one aoTuV b5, which is good but without a release for almost a year. LAME had always been releasing betas and reaching one more final release early next year.

Lame 3.98 beta 6

Reply #127
Vorbis I is not even the recommended Vorbis encoder in HA. All spotlight is directing to the only one aoTuV b5, which is good but without a release for almost a year. LAME had always been releasing betas and reaching one more final release early next year.


Ahem, the aoTuV versions are "just" patches to Vorbis I and these patches are now in the Vorbis I development branch...

Yes, lame releases one version each year but most important for these releases are sound quality improvements (if they happen). Lame 3.98 --vbr (especially the newest betas b5 and b6) is a true gem in that respect but between 3.90.3 and 3.97 I couldn't see that many improvements. In fact, 3.97 --vbr-new sounded quite bad to me (I was angry about -V2 several times) ) and I got the impression that --cbr and 3.90.3 is still a very good alternative. Now, how old is 3.90.3? It is from May 2003!

Lame 3.98 beta 6

Reply #128
In your opinion, would you make the Vorbis I encoder the recommended encoder since it has been released a few weeks ago and has the aoTuV patches? Or is it only in development branch? If not why aoTuV keeps being the recommended encoder even with low level noise added to encoded files?

LAME 3.98b6 should be set as the default encoder now.
3.97b1/2 was the recommended encoder for a long while (no point saying HA likes only "final" releases as stated by a member).

 

Lame 3.98 beta 6

Reply #130
Sorry guys, I haven't been here for ages and I don't have time to read through all the posts...

I haven't used MP3 since 3.96.1 final but I bought my mom a new MP3 player and so I needed to grab her a few CDs.
I googled for "lame", picked the first download, it was version 3.98b6 (oct 22 2007). Then I accidentally listened to one of the ripped songs and it was a shock for me!

here is the sample
http://www.volny.cz/martel/mp3test/test.zip

I used lame -b 128 -h command line to encode the song. The 398b6 version produced real nasty mess in the beginning:
http://www.volny.cz/martel/mp3test/test398.mp3
I somewhat remembered mp3 quality over those years and this was too much. So I used the 3.96.1 version I had on disk and encoded with the same command line:
http://www.volny.cz/martel/mp3test/test3961.mp3
And here it goes - the mess was gone!

I don't care much about MP3 development anymore but imagine a normal user which googles for Lame and gets this junk, far inferior compared to several years old version!

I was like "WTF?!!!" myself. There should really be some warning attached to beta releases or the code should be somewhat restricted so normal people googling for LAME get the best version available, not some hocus-pocus junk like the version i got.

EDIT: Sorry, the links are down as my hosting got deleted without notice. Unfortunately, I don't have those files anymore.

Lame 3.98 beta 6

Reply #131
Some regression is normal as long as overall quality is showing improvement. Also is CBR going through much tuning  / testing ? I guess most people are interested in the vbr scale. Also -h is q2 and is not the default noise shaping - maybe different from 3.96 q2. Sometimes I got increased ringing from it.
wavpack 4.8 -b256hx6c

Lame 3.98 beta 6

Reply #132
... it was version 3.98b6 (oct 22 2007) ....

Unfortunately premature and incorrect versions of 3.98b6 were compiled and published.
The 'real' 3.98b6 version was published december 17th and can be downloaded from Rarewares.
I guess everything is alright with this version, and you're welcome to try it.
lame3995o -Q1.7
opus --bitrate 140

Lame 3.98 beta 6

Reply #133
I used lame -b 128 -h command line to encode the song. The 398b6 version produced real nasty mess in the beginning:
http://www.volny.cz/martel/mp3test/test398.mp3
I somewhat remembered mp3 quality over those years and this was too much. So I used the 3.96.1 version I had on disk and encoded with the same command line:
http://www.volny.cz/martel/mp3test/test3961.mp3
And here it goes - the mess was gone!


Can't ABX these two. They sound identical to me.

Lame 3.98 beta 6

Reply #134

... it was version 3.98b6 (oct 22 2007) ....

Unfortunately premature and incorrect versions of 3.98b6 were compiled and published.
The 'real' 3.98b6 version was published december 17th and can be downloaded from Rarewares.
I guess everything is alright with this version, and you're welcome to try it.

If you mean this file
http://www.rarewares.org/dancer/dancer.php?f=186
then know that I tried it out now and it produces exactly the same mess as the previously mentioned version (oct 22).


I used lame -b 128 -h command line to encode the song. The 398b6 version produced real nasty mess in the beginning:
http://www.volny.cz/martel/mp3test/test398.mp3
I somewhat remembered mp3 quality over those years and this was too much. So I used the 3.96.1 version I had on disk and encoded with the same command line:
http://www.volny.cz/martel/mp3test/test3961.mp3
And here it goes - the mess was gone!


Can't ABX these two. They sound identical to me.

No offense meant, but I sincerely hope that you or people like you don't help develop LAME.
If you don't know what to look for - the 398 version adds audible (pre?)echo to both audio channels (before vocals come in). 3961 adds only almost inaudible one.
I use Sennheiser HD 490 headphones and an integrated Realtek HD audio sound card (Asus P5K). And I can hear the mess despite the mediocre setup I have.
If noone else can confirm the fuzz, then there's no hope for you guys. LOL
I think I'm going to back up my 3961 version of Lame pretty good.

Lame 3.98 beta 6

Reply #135
I haven't used MP3 since 3.96.1 final but I bought my mom a new MP3 player and so I needed to grab her a few CDs.

This new mp3 player should be able to play VBR mp3s. Try -V5 instead of -b 128 -h

Lame 3.98 beta 6

Reply #136
I haven't used MP3 since 3.96.1 final but I bought my mom a new MP3 player and so I needed to grab her a few CDs.

This new mp3 player should be able to play VBR mp3s. Try -V5 instead of -b 128 -h

No offense meant, but CBR (128, 160 and 192) is bread and butter of MP3. Is there really ANY reason to wreck it up in newer releases of LAME? I don't care about VBR as long as there is no two-pass encoding. And even if there was, I wouldn't trust a version that can't do CBR properly.

Lame 3.98 beta 6

Reply #137
128k cbr is the most archaic thing and was notorious for giving mp3 a stigma. IMO its too unstable no matter what encoder. You need at least 160 or better yet 192 for steady cbr.
wavpack 4.8 -b256hx6c

Lame 3.98 beta 6

Reply #138
[/quote]Can't ABX these two. They sound identical to me. [/quote]



Same here. No "mess" in my ears, as well.





@Martel



Most probably your Mother won't hear the "mess" either. It must be a real curse and a burden to go through life with your incredible hearing. 





Happy New Year!
Surf's Up!
"Columnated Ruins Domino"

Lame 3.98 beta 6

Reply #139
Martel, beta is beta, you need to use 3.97 final.

Lame 3.98 beta 6

Reply #140
I can confirm that the problem exists, I can ABX it without trouble. Will try the official 3.98b6 now to see if the problem still exists.

Edit: It's still a problem with the Dec 17 build. In fact, the created file is bit-identical with the problem file provided by Martel. The most obvious thing I can make out is a stereo problem; you can hear added stuff coming from the right in the 3.98 encode. Hard to describe this more clearly, sorry.

Edit2: An interesting point maybe, lame -V6 -b128 -B128 -F (with the Dec 17 build, also tried V0 with similar result) sounds better to me than any of the -b128 versions I listened to (i.e. also better than the "unproblematic" 3.96.1 encode). This is not encoding advice - it just seems to me that the VBR code can (at least sometimes) be better at cbr than the cbr code, which is kind of weird...

Lame 3.98 beta 6

Reply #141
Martel is correct. The problem is obvious. I can hear the phenomenon easily with my crappy PC speakers. No need to switch to headphones or to my Hi-Fi system. This is something completely different than a minor artifact that can be reliably detected only by ABX testing.

The same problem exists already in the official v. 3.97. I didn't have 3.96.1 anywhere near but I tried the old reference 3.90.3 and it is clean similarly like Martel's 3.96.1 sample.

Here is a screenshot of Spectrum Laboratory's waterfall display. It contains the  first three guitar chords (about 3 s.). The additional artifacts can be clearly seen in the 3.97 sample. The problem is more obvious in the left channel. The samples are (from top to bottom):

1. LAME 3.97 "release" @ -b 128  (which should use the optimized CBR presets automatically)
2. LAME 3.90.3 @ -b 128 -h (-m j is enabled by default)
3. original wave


Click to enlarge.


EDIT
The most obvious thing I can make out is a stereo problem; you can hear added stuff coming from the right in the 3.98 encode. Hard to describe this more clearly, sorry.

Could you have the channels reversed? Looks like there's more additional stuff on the left.

Lame 3.98 beta 6

Reply #142
Could you have the channels reversed? Looks like there's more additional stuff on the left.


Might be, but there's already the guitar there. It's easier to spot something when it comes from a previously "silent" direction I'd say.
Edit: I just listened to it with my speakers instead of headphones for the first time, now the distortion on the left is far more noticeable.

Lame 3.98 beta 6

Reply #143
...If you mean this file
http://www.rarewares.org/dancer/dancer.php?f=186
then know that I tried it out now and it produces exactly the same mess as the previously mentioned version (oct 22).....

Yes, that's the link I meant. And like Med0 and AlexB I can confirm your finding: the -b128 encoding is real bad, kind of a long-lasting additional echo.
No offense meant, but CBR (128, 160 and 192) is bread and butter of MP3. Is there really ANY reason to wreck it up in newer releases of LAME? I don't care about VBR as long as there is no two-pass encoding. And even if there was, I wouldn't trust a version that can't do CBR properly.

I agree with you that CBR should be fine as well (at least with this sample), and it seems to me too that general focus is too much on VBR in recent years. But I do think the Lame devs are willing to improve on problems found also with CBR. So your sample is valuable especially as not many people use CBR.
But shadowking's comment is true as well IMO: when using CBR a higher bitrate is necessary to get at real good results with mp3.
You can also see it the other way around: Lame 3.98b6 has become real good at VBR, and it's not a bad idea to use it.
Anyway it's up to you, and if you want to stick with CBR and want to use it right now, you may be better off using 3.96.1 or even 3.90.3.
lame3995o -Q1.7
opus --bitrate 140

Lame 3.98 beta 6

Reply #144
@Martel

Most probably your Mother won't hear the "mess" either. It must be a real curse and a burden to go through life with your incredible hearing.

I'm pretty sure my mother can hear the problem and she was born in the '30s. 
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Lame 3.98 beta 6

Reply #145
@Martel

Most probably your Mother won't hear the "mess" either. It must be a real curse and a burden to go through life with your incredible hearing.

I'm pretty sure my mother can hear the problem and she was born in the '30s. 


You know, I gotta re-listen to these samples again when I get home from work, because I coudn't hear it on these PC speakers. When Martel used the word "mess", I imagined this ungodly unlistenable noise, and even if there is this echo that you all hear is it that terrible that it makes it unlistenable?  That LAME is useless because of it? Do I need to get a doctor's referral for a hearing test? My 50 year old ears just don't hear the "mess" that Martel does.
Surf's Up!
"Columnated Ruins Domino"

Lame 3.98 beta 6

Reply #146
...even if there is this echo that you all hear is it that terrible that it makes it unlistenable?  That LAME is useless because of it?

If Lame is regressing (the sample sounds fine using 3.96 and 3.90) then it should be brought to the attention of the developers.
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

Lame 3.98 beta 6

Reply #147
... I imagined this ungodly unlistenable noise, and even if there is this echo that you all hear is it that terrible that it makes it unlistenable?  That LAME is useless because of it? Do I need to get a doctor's referral for a hearing test? My 50 year old ears just don't hear the "mess" that Martel does. ...

It's personal judgement how to feel about a problem.
Anyway it's a real problem that hopefully will be repaired.
lame3995o -Q1.7
opus --bitrate 140

Lame 3.98 beta 6

Reply #148
No offense meant, but CBR (128, 160 and 192) is bread and butter of MP3. Is there really ANY reason to wreck it up in newer releases of LAME? I don't care about VBR as long as there is no two-pass encoding. And even if there was, I wouldn't trust a version that can't do CBR properly.

No, there's no reason to wreck CBR quality and I hope this problem gets corrected. But why not care about VBR just because there is no two-pass encoding? And even if it butchers all samples with CBR, if it sounds better with VBR and works with your MP3 player then why not use it?


 
SimplePortal 1.0.0 RC1 © 2008-2020