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: Current status of MP3 encoders (Read 3076 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Current status of MP3 encoders

  Hi.

I found the results of this public test interesting. https://listening-tests.hydrogenaud.io/sebastian/mp3-128-1/results.htm
The test is dated but MP3 encoders haven't changed much since then.

Different MP3 encoders perform better on different type of music, different type of signals:
Helix is good at electronic music, sharp transients and good portion of instrumentals.
LAME 3.97 is good on rock and metal (where Helix doesn't perform well)
Fraunhofer is good at voice & singing parts.

These encoders keep high quality while not jumping  that much in bitrate, so it's not just bumping bitrate over all samples, but being smart on VBR algorithms and increasing bitrates where it is only required  ;)

It's a good indicator that there is still room to improve MP3. How much?  I would estimated it as equivalent to 15% of bitrate gain at 130-145 kbps, V5-V4 settings, looking at performance of different encoders from that public test.  It's been a long time and developers now have more tools to make a better encoders ( like better metrics to measure audio quality, faster hardware etc.)
Look at JPEG, modern encoders of old format as Guetzli squeezes extra 10-20% of quality using more advanced and complex psychovisual model.

Does anybody know if there is active development of any MP3 encoder?  There were some changes for Helix encoder this year but it's related to usability not quality. LAME 3.99 is faster than 3.98  but  pretty the same in terms of quality. And LAME 3.100/3.101 didn't bring any quality improvements.

Re: Current status of MP3 encoders

Reply #1
AFAIK, currently there's nothing serious to do with MP3. I think it could be even better to work in Musepack or Opus instead.
(but actually no idea of nothing :D)


Okay no. But I'm not sure of how far we can go. There's Vorbis and AAC. Different settings and encoders give different results, but actually there isn't one that provides an incredible increase of quality.

Re: Current status of MP3 encoders

Reply #2
Look at JPEG, modern encoders of old format as Guetzli squeezes extra 10-20% of quality using more advanced and complex psychovisual model.
If you look at JPEG, jpegli is even better; Without the Guetzli's more than 1000x slowness, and unlike Mozjpeg, jpegli is tuned for all ranges of quality, including high quality settings. You can test at
jxl-x64-windows-static.zip, cjpegli.exe.

Does anybody know if there is active development of any MP3 encoder?  There were some changes for Helix encoder this year but it's related to usability not quality. LAME 3.99 is faster than 3.98  but  pretty the same in terms of quality. And LAME 3.100/3.101 didn't bring any quality improvements.

As far as I know, there is no active development of MP3 encoders in 2024, outside of Resurrecting/Preserving the Helix MP3 encoder.

Re: Current status of MP3 encoders

Reply #3
Spot on about the need for another MP3 encoder or a lame 4.0 ig... MP3 may be old, but it's still the defacto at a lot of places on earth... a newer encoder sure won't hurt

The development ceased a long time ago, but there's still like 5-10% scope imho, for the "perfection" part

AAC at CVBR 256 easily beats MP3 V0 VBR, and I don't think it's due to a limitation of the format, but an encoder/model thing, which can be perfected a bit more

I mean... people comparing the ancient Helix/FhG to be better on certain music, speaks a bit about the possibilities ig

Re: Current status of MP3 encoders

Reply #4
AAC at CVBR 256 easily beats MP3 V0 VBR, and I don't think it's due to a limitation of the format,

If there weren't limitations of the format, there wouldn't be a need for new one, then :) but can you explain this "aac easily beats mp3" part... can you prove it? These are mighty high bitrates for two mature codecs.
TAPE LOADING ERROR

Re: Current status of MP3 encoders

Reply #5
"easily" is a very elastic concept. But some trained listeners ... https://hydrogenaud.io/index.php/topic,121579
Note, this is just one test, and too few samples to make it statistically significant.
(There are a couple of different ways to interpret the data - are they on a cardinal scale or not? But in both cases, the chance that random outcomes would give just as uneven race between AAC-192 and MP3-V0, is a bit too big to conclude. Of course, if you take into account all the other data, they say it is unreasonable to believe they are all from the same distribution.)

This was AAC at 192 - which doesn't imply that AAC at 256 would win so much more often that it would establish significance. But it gives a fairly good indication.

Re: Current status of MP3 encoders

Reply #6
"easily" is a very elastic concept. But some trained listeners ... https://hydrogenaud.io/index.php/topic,121579
Note, this is just one test, and too few samples to make it statistically significant.
(There are a couple of different ways to interpret the data - are they on a cardinal scale or not? But in both cases, the chance that random outcomes would give just as uneven race between AAC-192 and MP3-V0, is a bit too big to conclude. Of course, if you take into account all the other data, they say it is unreasonable to believe they are all from the same distribution.)

This was AAC at 192 - which doesn't imply that AAC at 256 would win so much more often that it would establish significance. But it gives a fairly good indication.

Thanks for that porcus, we got some numbers to prove my point, even if it isn't the 1000 people testing standard of hydrogen audio to consider anything 'signifcant'..

however that 4.45 vs 4.92 of AAC at 192kbps sure seems worthy to look into, especially when people here claim to say that at bitrates above 160 all encoders are good enough... and at 192/256 you're splitting hairs with any encoder..

I mean ymmv but most young listeners with good setups can differentiate 192 and 256, and definitely AAC from MP3

Thanks again

Re: Current status of MP3 encoders

Reply #7
most young listeners with good setups can differentiate 192 and 256, and definitely AAC from MP3
Is there substantiation to that claim?  Just being picky, but "most" means more than 50%, and where's your sample?
It's your privilege to disagree, but that doesn't make you right and me wrong.

Re: Current status of MP3 encoders

Reply #8
I noticed there is commonly a confusion arising from forgetting that when you compare versions of a song in different formats that one didn't encode themselves (from a common source), then the observable difference is actually more likely to be from dozens of other factors (different mastering, additional processing in a proprietary player or radio software, one or both versions using an ancient and/or sub-par encoder that was poorly tuned, generation loss when not using a clean source, etc…)  and not just the codec that happened to be used at the end for these song versions.
a fan of AutoEq + Meier Crossfeed

Re: Current status of MP3 encoders

Reply #9
"easily" is a very elastic concept. But some trained listeners ... https://hydrogenaud.io/index.php/topic,121579
Note, this is just one test, and too few samples to make it statistically significant.
This was AAC at 192 - which doesn't imply that AAC at 256 would win so much more often that it would establish significance. But it gives a fairly good indication.
Pardon, but aren't these those killer, hard-to-encode samples, and not real life music?
If yes, I'd like to see someone differentiate before mentioned aac and mp3 music, to see how they would fare. I'm old, tho, but in my ABX testings 15-20 years ago I couldn't tell a difference at bitrates above 160 kbit.

It might be time for organizing public ABX testing, we do have some new codecs.
TAPE LOADING ERROR

Re: Current status of MP3 encoders

Reply #10
"easily" is a very elastic concept. But some trained listeners ... https://hydrogenaud.io/index.php/topic,121579
Note, this is just one test, and too few samples to make it statistically significant.
This was AAC at 192 - which doesn't imply that AAC at 256 would win so much more often that it would establish significance. But it gives a fairly good indication.
Pardon, but aren't these those killer, hard-to-encode samples
"elastic concept" I said  ;)

Let's say "is 16 bits better than <suitable bit depth N>?", where N is so high that for a ton of music, you won't hear the difference - yet N is so low that it is not hard to find music where you can measure that trained ears hear the difference.
Easily ... ?
If the bar is "random sample", no. If the bar is "not hard to find", then yes.
 
I'd say, when we are up to > 200 kbit/s, then the yardstick is more like the latter ... ?

Re: Current status of MP3 encoders

Reply #11
Pardon, but aren't these those killer, hard-to-encode samples, and not real life music?
These were indeed hard-to-encode samples and at the same time they are real life music. Nos synthetic frequency sweeps.  Still, these were killer samples for MP3 (at least for V2) but not so for AAC.   ;)
Still I agree with you that "aac easily beats mp3" at 256 kbps is exageration.  It's well accepted here that LAME V2 (192 kbps) produces transparent results for wide public.  So it's ok if some  well trained listeners like Kamedo2, IgorC, Guruboolez can perform blind tests at higher bitrates. 

Yet there is strong evidence that AAC 96 kbps outperforms MP3 128 kbps  https://listening-test.coresv.net/results.htm
Kamedo2 tests also confirm that AAC 192 is better than LAME MP3 192 kbps, but again he is extremely trained listener.
https://kamedo2.hatenablog.jp/entry/20111029/1319840519


It might be time for organizing public ABX testing, we do have some new codecs.
Where?
I can see only two xHE-AAC encoders (exhale and Poikosoft/FhG) and their popularity is still low.
  • MP3 - Lame/Helix. No quality changes since last public tests
  • AAC - Apple is still the best AAC encoder already for 20 years
  • Aotuv Vorbis, Musepack - no change
  • Opus - no change, except speech quality at very low bitrate, I don't think  here folks will be too excited to test it at that low bitrate.

 

Re: Current status of MP3 encoders

Reply #12
Pardon, but aren't these those killer, hard-to-encode samples, and not real life music?
These were indeed hard-to-encode samples and at the same time they are real life music. Nos synthetic frequency sweeps.  Still, these were killer samples for MP3 (at least for V2) but not so for AAC.   ;)
Still I agree with you that "aac easily beats mp3" at 256 kbps is exageration.  It's well accepted here that LAME V2 (192 kbps) produces transparent results for wide public.  So it's ok if some  well trained listeners like Kamedo2, IgorC, Guruboolez can perform blind tests at higher bitrates. 

Yet there is strong evidence that AAC 96 kbps outperforms MP3 128 kbps  https://listening-test.coresv.net/results.htm
Kamedo2 tests also confirm that AAC 192 is better than LAME MP3 192 kbps, but again he is extremely trained listener.
https://kamedo2.hatenablog.jp/entry/20111029/1319840519


It might be time for organizing public ABX testing, we do have some new codecs.
Where?
I can see only two xHE-AAC encoders (exhale and Poikosoft/FhG) and their popularity is still low.
  • MP3 - Lame/Helix. No quality changes since last public tests
  • AAC - Apple is still the best AAC encoder already for 20 years
  • Aotuv Vorbis, Musepack - no change
  • Opus - no change, except speech quality at very low bitrate, I don't think  here folks will be too excited to test it at that low bitrate.


I'm not the kind to pick up my daw and start ABX'ing rn because I'm a bit too occupied to do that all for proving my point, but I can assure you, that a few years ago when I had not much idea about AAC/MP3 encoding or what I preferred (back then it was all about compatiblity and not necessarily "choice")

I always went with AAC except in situations where AAC won't run on the device... in my hifi system as well, my chain is transparent class AB, feeding from my foobar straight to the DAC exclusively, AAC always sounds better... just like the public tests confirm...

Apple AAC - the best

The point being, it's been 20 years since that encoder came, and more like 15 years since it has had no updates... and I find it very hard to believe that it can't be improved further with the tech we have today... it's just not reasonable yk?

PCM is decades old yes I know, but it's not a model, it's uncompressed... this is compression and psychoacoustic modelling... and it's kinda hard to believe that AAC or MP3 can't be improved further...

We may be 98% there, but there's still a 2% chance I believe in, especially for MP3... don't quote me on the 2% etc numbers, they're pulled out of air just to express my point, but ykwim..

We have AoTuV for Ogg... maybe something like this for LAME?

Re: Current status of MP3 encoders

Reply #13
The point being, it's been 20 years since that encoder came, and more like 15 years since it has had no updates... and I find it very hard to believe that it can't be improved further with the tech we have today... it's just not reasonable yk?
I don't understand where you're coming from.

MP3, AAC, whatever, have defined methods for pre-processing and encoding, then decoding and post-processing.  Modern tech makes it faster to perform those stages in software (or implement them directly in hardware), but it can't change the basic architecture of the processing stages otherwise it wouldn't be MP3/AAC.

Minor tweaks to the psycho-acoustic model won't yield the step improvement in "quality" you're looking for, and might result in a split opinion whether the tweak is actually better.  Major tweaks would be the equivalent of ripping it up and starting again – ie not MP3/AAC at all, a new codec.

Codecs are limited by the era they were created in.  In particular, the decoder had to be able to decode in real time on the hardware available in that era, which put a ceiling on the complexity of architecture which could be accommodated.  You can't now turn around and say "we have much faster hardware now, so let's redefine the architecture of MP3".

AAC is much more recent than MP3, therefore has the capacity to be much better than MP3, but I suggest that both have reached their ceilings (or very close to them).
It's your privilege to disagree, but that doesn't make you right and me wrong.

Re: Current status of MP3 encoders

Reply #14
If you look at JPEG, jpegli is even better; Without the Guetzli's more than 1000x slowness, and unlike Mozjpeg, jpegli is tuned for all ranges of quality, including high quality settings. You can test at
jxl-x64-windows-static.zip, cjpegli.exe.
Thank you for link to binaries, Kamedo2.

Here is visual comparison of fresh versions of mozgpeg vs jpegli vs avif, etc... jpegli is really good

MP3, AAC, whatever, have defined methods for pre-processing and encoding, then decoding and post-processing.  Modern tech makes it faster to perform those stages in software (or implement them directly in hardware), but it can't change the basic architecture of the processing stages otherwise it wouldn't be MP3/AAC.
It's still possible to improve encoders without breaking compatibility  with MP3 decoders, even  if format is old as JPEG or MP3.
jpegli (new JPEG encoder in 2024  :) ) is a good confirmation to that.  And JPEG is even older than MP3 format.
It's clear  that current status of JPEG vs MP3 for reaching format max capabilites is unknown, and  this topic is just that, to ask/discuss/estimate/inform. While JPEG and MP3 are two different formats, one is for image coding, another is for audio, still both  use some common techniques as MDCT.

Re: Current status of MP3 encoders

Reply #15
I suggest there is more room for manoeuvre with 2D than 1D.
It's your privilege to disagree, but that doesn't make you right and me wrong.

Re: Current status of MP3 encoders

Reply #16
The point being, it's been 20 years since that encoder came, and more like 15 years since it has had no updates... and I find it very hard to believe that it can't be improved further with the tech we have today... it's just not reasonable yk?
I don't understand where you're coming from.

MP3, AAC, whatever, have defined methods for pre-processing and encoding, then decoding and post-processing.  Modern tech makes it faster to perform those stages in software (or implement them directly in hardware), but it can't change the basic architecture of the processing stages otherwise it wouldn't be MP3/AAC.

Minor tweaks to the psycho-acoustic model won't yield the step improvement in "quality" you're looking for, and might result in a split opinion whether the tweak is actually better.  Major tweaks would be the equivalent of ripping it up and starting again – ie not MP3/AAC at all, a new codec.

Codecs are limited by the era they were created in.  In particular, the decoder had to be able to decode in real time on the hardware available in that era, which put a ceiling on the complexity of architecture which could be accommodated.  You can't now turn around and say "we have much faster hardware now, so let's redefine the architecture of MP3".

AAC is much more recent than MP3, therefore has the capacity to be much better than MP3, but I suggest that both have reached their ceilings (or very close to them).

See, it's not about redefining the architecture of the format, the file will be adherent to the same standard, it's the audio data, the encoding complexity that can be further increased, even if it's at the expense of time to yield better quality results

Re: Current status of MP3 encoders

Reply #17
Here is visual comparison of fresh versions of mozgpeg vs jpegli vs avif, etc... jpegli is really good
I'm using it for some conversions (it's in XNView) and the encoding speed is just not noticable, it's like it just copies files, very impressive. But from the result, HEIC impresses me more, it doesn't suffer from that ghosting at contrasting edges...
jpegli replaced webp as for Google Photos for me (pictures via Chromecast for 4K HDTV). The jpegs are in the 800k-1.2 MB range. 4k and good enough!

Re: Current status of MP3 encoders

Reply #18
Funny thing is that it was me with a different username that started a thread about jpegli at the Xnview forum that lead to its implementation. I once suggested the use and supported Xnview at a former employer.
However i don't use it myself because of such limitations: math shortcomings
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: Current status of MP3 encoders

Reply #19
The point being, it's been 20 years since that encoder came, and more like 15 years since it has had no updates... and I find it very hard to believe that it can't be improved further with the tech we have today... it's just not reasonable yk?
I don't understand where you're coming from.

MP3, AAC, whatever, have defined methods for pre-processing and encoding, then decoding and post-processing.  Modern tech makes it faster to perform those stages in software (or implement them directly in hardware), but it can't change the basic architecture of the processing stages otherwise it wouldn't be MP3/AAC.

Minor tweaks to the psycho-acoustic model won't yield the step improvement in "quality" you're looking for, and might result in a split opinion whether the tweak is actually better.  Major tweaks would be the equivalent of ripping it up and starting again – ie not MP3/AAC at all, a new codec.

Codecs are limited by the era they were created in.  In particular, the decoder had to be able to decode in real time on the hardware available in that era, which put a ceiling on the complexity of architecture which could be accommodated.  You can't now turn around and say "we have much faster hardware now, so let's redefine the architecture of MP3".

AAC is much more recent than MP3, therefore has the capacity to be much better than MP3, but I suggest that both have reached their ceilings (or very close to them).

The ironic thing is that MP3 decode complexity is actually relatively high by modern standards since you have to do both the MDCT and filterbank.  AAC-LC is quite efficient in that you just do a single transform.  Aside from the low bitrate extensions like SBR, newer codecs have not much changed the overall decoder complexity much since the 1990s, they've just used that complexity more cleverly. 

Re: Current status of MP3 encoders

Reply #20
Funny thing is that it was me with a different username that started a thread about jpegli at the Xnview forum that lead to its implementation. I once suggested the use and supported Xnview at a former employer.
However i don't use it myself because of such limitations: math shortcomings
Ugh, some painful and historic burden, it seems.

Re: Current status of MP3 encoders

Reply #21
What am I loosing by going aac-lc ?  Any technical drawbacks excluding a rare legacy device ?

Re: Current status of MP3 encoders

Reply #22
Replacing lossless? Since lossy is just as good as lossless at the right bitrate it is right to assume that you will be perfectly fi... dont do it.
And so, with digital, computer was put into place, and all the IT that came with it.

Re: Current status of MP3 encoders

Reply #23
What am I loosing by going aac-lc ?  Any technical drawbacks excluding a rare legacy device ?
Audio and file size - you're winning. Can only talk of my issues. I use LC-AAC mostly - but: My cars don't display embedded album art, hence I use mp3 there.