LAME pushes MP3s further with the upcoming 3.89
detail information
| Posted by | Robin van Lieshout |
| Posted on | 25/05/01 19:12 |
Copyright 2001 CD-RW.ORG
Click for lowest prices in CD-R media
LAME pushes MP3s further with the upcoming 3.89
WARNING: LAME 3.89 is at alpha stage. Even though LAME alphas do not crash or cause similar trouble, their audio features may change from day to day. Last approved version is LAME 3.88b.
LAME, the ultimate MP3 encoder, brought MP3 to a whole new level of quality with the 3.88 version. It took months for LAME to update from 3.87 to 3.88, but since the 3.88 the development pace has been furious. Biggest changes implemented so far is the new default VBR mode which provides about 100% speed improvement and improved Joint Stereo algorithms.
For some reason the MP3 scene likes to stick with the old Fraunhofer encoder with 192kbps/stereo settings. This amazes me since LAME encoder has outperformed the Fraunhofer ages ago, somewhere around the 3.6 - 3.7 versions-that is well over a year ago. Equally amazing is that the scene continues to stick with 192kbps encoding, while all modern encoders utilize VBR for quality improvements. The anti-VBR attitude must originate from the days of the XING encoder, which did produce terrible results. But since then times have changes and if you look at the LAME Encoder changelog you'll realise the amount of work put to the optimisation of this fabulous codec. With LAME VBR it is possible to achieve virtually perfect quality with average bitrates around 170-210 kbps.
Now back to the version 3.89
The new VBR routines of 3.89 are 2x faster than algorithms used in the previous versions. However it's known to have some quality issues vs. the old routines, but currently plenty of development effort is put to the quality of the new VBR. And note that the new VBR already produces quality over all other encoders, but the older (slower) routines may just be even tiny bit better. Based on my personal tests, the new VBR is already at very high level of quality and shouldn't produce problems in normal use. However, high-end extremists may want to stick with old VBR that is still available also.
The MP3 scene likes to use stereo encoding, while Joint Stereo provides significant quality improvements. Even the Fraunhofer recommends Joint Stereo below 256kbps bitrates.
I'll try to explain Joint Stereo very briefly. Basically it looks for signals that are identical in the left and right channel and if it finds any they are encoded as mono. This means that 50% bits are saved for the mono encoded signals and these bits are used to improve the encoding accuracy (very simplified explanation). If think rap music for example, most of the music is mono: the beat, the bass tunes, most of the vocals. Then there are some instruments & effects that come from left or right speaker. When you think of this, you can understand how much bits can be saved and encoding improved with Joint Stereo.
The JS is not problem free though. No algorithm is perfect and it may make mistakes. Once again the LAME Encoder has the most accurate JS available and the 3.89 shows significant changes again. If you use Guerillasoft's neat EncSpot ( www.guerillasoft.com ) , you'll see that the 3.89 version uses more strict stereo encoded frames than older LAME revisions. There is no documentation yet about the changes, but this clearly implies to ever higher accuracy of the JS algorithms.
Currently LAME Joint Stereo should be used to encode any studio recorded material. Studio recording have usually very strict stereo channel mastering, which means that sounds come clearly from both of the speakers, or strongly from the left or right. It's unlikely that the JS algorithms would make an error when encoding studio material. However, with music that is recorded in open space with multiple microphones (classical, some acoustic guitar, some live recordings..) stereo encoding may be a better choice, since in such recording the sounds don't come strictly form left, right or centre, but the music 'lives" somewhere in between the two channels. This may confuse the JS algorithms.
How to encode?
LAMEs best and the worst feature is the almost unlimited amount of different settings and options. Newbies will find it confusing, but experts love the possibilities to tweak the quality to exactly meet their needs at the moment. One can make special lower quality settings for his MP3 walkman and higher quality settings for his Hi-Fi equipment.
To play it safe and get excellent quality VBR MP3s you should use LAME 3.88b and the thoroughly tested settings:
--r3mix -b 128
These settings can easily be added to most of the CD rippers, while Exact Audio Copy is strongly recommended for best quality. Bitrates with these settings vary a lot: rap may encode at the average of 160kbps and heavy metal around 250kbps. If you listen music cross over, you should get an average below 200kbps in your encodings.
If you want to goof around with the new 3.89 alpha features, then you may try parameters like:
--adapt-thres-type 2 -V 1 -q 0 -m j --lowpass 19.5 --athtype 3
You may want to try V values of 0 or 2 and athtype of 2. These parameters should give you similar quality than the 3.88 settings above, but with 2x speed improvement. The bitrates with these settings vary less than with 3.88b, and they move around 170-210 kbps. Keep in mind that the new VBR is NOT completely tested or even finished yet.
It's impossible to write a complete LAME manual or tutorial within one article. Therefore here are some references you should look at for more information:
- http://cd-rw.org/forum
The highest quality MP3 community in the WWW. Many LAME developers and other experts are around. Also CD-R, DVD and other issues.
- http://www.r3mix.net
The original LAME-site, with very good tutorials and quality analysis.
- http://www.hot.ee/smpman/mp3/
Good place to download latest LAME versions
- http://www.mp3dev.org/mp3/
The LAME homepage.
Copyright 2001 CD-RW.ORG
Want to submit your own review? Click here
Posted by stvastva on Friday 25 May 2001 19:53
Excellent article! Thanks for all the specific information. :-)
The article is behind on reality; LAME 3.89a(25/05) VBR is now way better than that of any previous release. The 'quality issues' the writer of this article speaks about are now only positive bitrate (over)calculations, i.e. they can never deliver lower quality MP3 than older VBR routines. Despite of the author's personal testing (of which we've never seen any results) 'high-end extremists' would rather go with the new VBR than still use the old one.
Also, the official binary site for distributions of Win32 LAME files is not the one by smpman mentioned in the links,
it is the site by Dmitry Kutsanov:
http://dkutsanov.chat.ru
Other good links would be:
http://jthz.com/mp3/
the mp3-board:
http://r3.ma.cx
and the mailinglist:
http://lame.ma.cx
greetings,
JT + HZ
Also, the official binary site for distributions of Win32 LAME files is not the one by smpman mentioned in the links,
it is the site by Dmitry Kutsanov:
http://dkutsanov.chat.ru
Other good links would be:
http://jthz.com/mp3/
the mp3-board:
http://r3.ma.cx
and the mailinglist:
http://lame.ma.cx
greetings,
JT + HZ
Posted by LjPp on Saturday 26 May 2001 00:30
JT + HZ,
Unfortunately the new VBR routines are KNOWN TO FAIL with certain test samples in listening comparisons, when compared to the VBR old. So the older routines still produce better quality. (Naturally, there may be exceptions to this)
LAME developers now have the speed, but the real task is tweaking the new VBR to the level of the old routines.
See our board http://cd-rw.org/forum and the section "Technical tests & experiments"
Unfortunately the new VBR routines are KNOWN TO FAIL with certain test samples in listening comparisons, when compared to the VBR old. So the older routines still produce better quality. (Naturally, there may be exceptions to this)
LAME developers now have the speed, but the real task is tweaking the new VBR to the level of the old routines.
See our board http://cd-rw.org/forum and the section "Technical tests & experiments"
Posted by Redneck on Sunday 03 June 2001 00:21
Release groups should use Lame :: Yes
Reason :: superiour to Fraunhofer
Release groups should use VBR :: NO
Reason :: No more indication of quality possible, a vbr (with values between 56kbps and 192kbps) is a vbr, whilst a 192 is a 192kbps and a 56 is a 56kbps
Reason :: superiour to Fraunhofer
Release groups should use VBR :: NO
Reason :: No more indication of quality possible, a vbr (with values between 56kbps and 192kbps) is a vbr, whilst a 192 is a 192kbps and a 56 is a 56kbps
Posted by LjPp on Sunday 03 June 2001 22:57
VBR typically moves around 128-320kbps.
Indicators for quality:
-Your ears
-Winamp bitrate display
-EncSpot (great, check it out! www.guerillasoft.com
Indicators for quality:
-Your ears
-Winamp bitrate display
-EncSpot (great, check it out! www.guerillasoft.com
The reason the scene doesn't like joint stereo is flanging. Yeah, js makes files smaller and keeps the quality ok, but the bottom line is quality.
And another thing. Speed for most of us, I daresay is not a factor. It's a few of us encoding for _many_ of them. They don't care about speed of encoding. I don't either. I would rather have it take twice as long, if I can get a 10% increase in quality. It's the quality of your work that you will be remembered by, or does nobody remember that from the old days?
Posted by A_MEN on Thursday 16 August 2001 12:05
Good quality is a must if u have good speakers. But with those bad quality speakers uget no difference between 128 and 256 kbps
I'm kind of tired of the newbies trying to tell us which encoder is best based on some lame (heh) tests with sinus wave graphs..
Let me tell you one thing: for an experienced ripper, FHG is still the way to go (and updated versions are available, just radium is too lazy to rip them out). LAME is still unable to encode difficult examples that are prone to flanging, unless it is used at very high bitrates.
The example I use did not encode any better with new versions.. I still try the latest out of CVS every now and then but this doesn't change much.
In regards to overall quality of encoding with _a large number of rips_, FHG is still the best and safest, whether it sounds less "crispy" here and there or not!
The real advantage is of course the open source code that makes it possible to learn from it, and a 20-30% speed increase by compiling with optimizations is not bad either.
But other than that, LAME still has to catch up.
It's not good if people believe the hype more than their own ears. On the other hand, geek ears are notoriously deaf for these things, and hardly anyone spends a 4 digit $$ price for his speakers as it would be needed.
So support the hype as much as you want, your articles and the meaningless frequency response graphs will NOT make the ringing and hissing go away.
Let me tell you one thing: for an experienced ripper, FHG is still the way to go (and updated versions are available, just radium is too lazy to rip them out). LAME is still unable to encode difficult examples that are prone to flanging, unless it is used at very high bitrates.
The example I use did not encode any better with new versions.. I still try the latest out of CVS every now and then but this doesn't change much.
In regards to overall quality of encoding with _a large number of rips_, FHG is still the best and safest, whether it sounds less "crispy" here and there or not!
The real advantage is of course the open source code that makes it possible to learn from it, and a 20-30% speed increase by compiling with optimizations is not bad either.
But other than that, LAME still has to catch up.
It's not good if people believe the hype more than their own ears. On the other hand, geek ears are notoriously deaf for these things, and hardly anyone spends a 4 digit $$ price for his speakers as it would be needed.
So support the hype as much as you want, your articles and the meaningless frequency response graphs will NOT make the ringing and hissing go away.
Some guy wrote:
"Unfortunately the new VBR routines are KNOWN TO FAIL with certain test samples in listening comparisons, when compared to the VBR old. So the older routines still produce better quality."
Well, this is nonsense. Not only do we personally have yet to hear proof of that, also it is fact that LAME changes per day of new release. For example, your post was done on a day where the new VBR was way better than the old one already. Also, the test examples are not conclusive, FhG fails a lot on some test-samples, where LAME would not, and vice versa, but the overall range of better encoding certainly comes from LAME encoded MP3's, not from any other encoder out there. We use MP3 encoding daily as well as professionally, and we sure know what we're writing about. You can test as much as you like, I will bet you that LAME will win on the average dance-track.
"Unfortunately the new VBR routines are KNOWN TO FAIL with certain test samples in listening comparisons, when compared to the VBR old. So the older routines still produce better quality."
Well, this is nonsense. Not only do we personally have yet to hear proof of that, also it is fact that LAME changes per day of new release. For example, your post was done on a day where the new VBR was way better than the old one already. Also, the test examples are not conclusive, FhG fails a lot on some test-samples, where LAME would not, and vice versa, but the overall range of better encoding certainly comes from LAME encoded MP3's, not from any other encoder out there. We use MP3 encoding daily as well as professionally, and we sure know what we're writing about. You can test as much as you like, I will bet you that LAME will win on the average dance-track.
Related tags
A tag is a relevant keyword or term associated with or assigned to a piece of information (like picture, product, or video clip), thus describing the item and enabling keyword-based classification of information it is applied to. We use it to make searching our database easier and enable all our users to contribute to our database.
add a tag
Add related tags
Related reviews/articles
Get the latest reviews via RSS 
