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.

Poll

Are you interested in a Transcoding Listening Test

I am interested and I want to participate
[ 42 ] (22.5%)
I am interested but I don't want to participate
[ 92 ] (49.2%)
I am not interested
[ 53 ] (28.3%)

Total Members Voted: 220

Topic: Are you interested in a Transcoding Listening Test (Read 21991 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Are you interested in a Transcoding Listening Test

There are many different codecs that HA uses, and each of them have their pros and cons, but one thing that most of them lack is the widespread hardware support of MP3.  This problem often encourages us to either backup our audio collection into MP3 or lossless all the time.

Now, given that portable hardware does not require transparent MP3s and other codecs are encoding at transparent levels, the question following question must be asked: "Shouldn't non-transparent MP3s sound the same if they were transcoded from transparent sources?"

This is something I'm genuinely curious about.  If MPC --standard and --extreme are both transparent, then shouldn't all MP3s made from them sound the same?  The answer is obvious but the explanation is not.  I'm proposing a test of popular codecs at popular settings and having them transcoded to a popular mp3 setting. I know that Den has already conducted a test like this on ATRAC, but I'd like to see how it fairs with the general HA community (and maybe beyond...).

Things I would like to know from you guys is what codecs/settings should be included, and what should the target should be.  I had privately suggested to Rjamorim that the target be lame -aps -y, but he said that it may not be a good setting for this test.

Thoughts anyone?

Are you interested in a Transcoding Listening Test

Reply #1 –
If you'll exclude WavPack and OptimFrog, it will purely be perceptual codec transcoding into MP3. If you could also calculate the eaqual values (transcoded vs. original) I'd appreciate it. Why MP3 but not AAC ?
The object of mankind lies in its highest individuals.
One must have chaos in oneself to be able to give birth to a dancing star.

Are you interested in a Transcoding Listening Test

Reply #2 –
I was debating on whether WavPack and OptimFrog should be included or not.  Den felt that WavPack was the best to transcode from, but I was concerned that WavPack could be identified easily and it could potentially skew results.  However, I'd like to hear from those interested in this test, whether they'd like to see WavPack and OptimFrog included.

As for why MP3 and not AAC, I can tell you in one word.  Legacy.  Hardware players are starting to come out for AAC, but I don't think the switch over will happen overnight (if even at all).  I estimate that MP3 players will still be predominant for the next 3-5 years, and they will have at least equal market share for a year or two beyond that.

That all said, I'd like to conduct this test with the HA community in mind, so if there is a strong preference for it, I will include it.  I'd like to have only one or two targets if possible in order to keep the size of the test from being HUGE.

Are you interested in a Transcoding Listening Test

Reply #3 –
I am for it. I think a good target to transcode to would be LAME --preset medium, which, although doesn't have the widespread acceptance as aps or aps -Y, still performs quite decently for portable use. On the other hand, since it has a relatively low bitrate (~150 kbps in my experience), I think it might be more suitable than APS to show the particular artifacts introduced by transcoding from the source codec. Another good choice would be --alt-preset 128, just for the 128 kbps popularity... However, I still propose --preset medium.

~Dologan

Are you interested in a Transcoding Listening Test

Reply #4 –
An interesting idea, MPC is supposed to transcode very well to MP3, so this test could add some weight towards the argument.

I voted intersted but won't take part after my horrible failure in Roberto's test..  if I can't ABX Blade 128kbs most of the time, I feel I am of no use to you

edit:  I agree that --preset medium should be considered for the MP3 setting, it's the setting I use as standard is excessively large for my needs, and my 32mb portable
< w o g o n e . c o m / l o l >

Are you interested in a Transcoding Listening Test

Reply #5 –
Quote
As for why MP3 and not AAC, I can tell you in one word. Legacy. Hardware players are starting to come out for AAC, but I don't think the switch over will happen overnight (if even at all). I estimate that MP3 players will still be predominant for the next 3-5 years, and they will have at least equal market share for a year or two beyond that.


Hmm surely new AAC players will play MP3 files as well. They're not giving up on MP3 and I don't think they ever will. But I think the AAC support is already there (iRiver, iPod, Philips). So this is not an issue of switch. I don't have any mp3 files but have been considering transcoding some of my MPC music to AAC for portable use... I will keep the MPC originals surely.
The object of mankind lies in its highest individuals.
One must have chaos in oneself to be able to give birth to a dancing star.

Are you interested in a Transcoding Listening Test

Reply #6 –
Quote
I voted intersted but won't take part after my horrible failure in Roberto's test.. if I can't ABX Blade 128kbs most of the time, I feel I am of no use to you

I'd stick with it - I can happily listen to 128k xings all day, but ogg -q6 transcoded to ogg -q2.5 made me feel ill it was so rubbish.
Given the number of people who claim transcoded oggs sound better than the original, a proper listening test might be a good idea.

There are so many different reasons for transcoding though:
  • Reducing the bitrate to fit more songs into limited storage.
  • Converting an unsupported format to a supported format.
  • Unintentionally - eg ripping a CD which was burned from MP3s.
  • Complete gimps re-encoding 56k mp3s to 320k to 'improve' them.
  • Ripping a DVD to a DivX (AC3 --> MP3 / Ogg).
  • My portable (a Palm TT) doesn't support 48khz so I've had to temporarily re-encode oggs to 44.1khz.
  • etc etc etc.
Some of these are high bitrate --> low bitrate, some vice versa.
Some of these are transcodes to the same format, some to alternative formats.
Some of these are attempts to get the best possible fidelity whatever the bitrate, some of these are attempts to get the best possible bitrate while retaining acceptable fidelity.


I see a VERY big test coming up.

Are you interested in a Transcoding Listening Test

Reply #7 –
Quote
Hmm surely new AAC players will play MP3 files as well. They're not giving up on MP3 and I don't think they ever will. But I think the AAC support is already there (iRiver, iPod, Philips). So this is not an issue of switch. I don't have any mp3 files but have been considering transcoding some of my MPC music to AAC for portable use... I will keep the MPC originals surely.

I'm still skeptical, but since there IS rising support for it, I'll have one target be AAC.  But I'm going to need an appropriate setting for hardware use (preferably for use with a flash-based device, since space is VERY limited there).

I'll have a look at Lame -ap medium when I get home tonight.  I'm a tad skeptical of it though, if it provides ~150kbps.  That's because supposedly, Lame -aps -Y is ~160kbps and is nearly transparent.  10kbps is borderline negligable.  If the difference is greater than that though, I'll seriously consider it.

Lame -ap 128 seems reasonable though, since most companies advertise how many songs their product can hold in terms of 128kbps mp3s.

Quote
If you could also calculate the eaqual values (transcoded vs. original) I'd appreciate it


.... er, how do I go about doing that?

Are you interested in a Transcoding Listening Test

Reply #8 –
I'm interested especially in mpc standard / xtreme -> mp3 apfs. That's what I do all the time and for portable use it has worked quite well. If you have a cd-mp3-player, why settle for less than ap(f)s? Most settings lower than aps are prone to artifacting even when encoding from the original wav. What should be tested here is allways the direct encode vs the transcode. Never the original vs the transcode.

Transcoding from ogg vorbis q5/6 wouls also be interesting. You could throw in Wavpack or Optimfrog hybrid, but I think they are superior by design (but who knows). Anyway, they are kinda niche (but maybe they would gain wider acceptance if their transcoding superiority would be proven in a test).

Personally I don't care about AAC as a target format.

Are you interested in a Transcoding Listening Test

Reply #9 –
Quote
.... er, how do I go about doing that?


You can find the software in the Rarewares section...
The object of mankind lies in its highest individuals.
One must have chaos in oneself to be able to give birth to a dancing star.

Are you interested in a Transcoding Listening Test

Reply #10 –
Quote
I'm interested especially in mpc standard / xtreme -> mp3 apfs. That's what I do all the time and for portable use it has worked quite well. If you have a cd-mp3-player, why settle for less than ap(f)s? Most settings lower than aps are prone to artifacting even when encoding from the original wav. What should be tested here is allways the direct encode vs the transcode. Never the original vs the transcode.

mpc -standard ABSOLUTELY has to be in this test.  I'll accept xtreme as well.  As for the Vorbis, q6 ought to be in there.  Would anyone object if it was a GT3 q6, rather than a straight Xiph compile?

q5, I'm not as sure on.  Anyone else have thoughts for or againts q5?

As for the WavPack and OptimFrog hybrids, is there anyone who would REALLY like to see either of these in the tests?  The reason I is because the fewer codec sources we have, the more target variations we can have.

Are you interested in a Transcoding Listening Test

Reply #11 –
Problem of any listening test : you have to select the targeted bitrate (and/or audioformat). Here, you have to choose the input too (means : format AND setting)... Good luck if you want to be ecumenical 

Personnal, I'm mainly interested by mpc as input, but at --insane profile or above. As output : lame --medium, AAC QT 128, AAC Ahead --streaming and WMApro 128 two-pass.

Are you interested in a Transcoding Listening Test

Reply #12 –
Quote
Problem of any listening test : you have to select the targeted bitrate (and/or audioformat). Here, you have to choose the input too (means : format AND setting)... Good luck if you want to be ecumenical 

Personnal, I'm mainly interested by mpc as input, but at --insane profile or above. As output : lame --medium, AAC QT 128, AAC Ahead --streaming and WMApro 128 two-pass.

The goal with the input is to have formats that an HA regular would have in his music archive.

If everyone interested just wants to see how MPC fairs against itself, then we can do that (hell, that sounds pretty damn tempting to me).  I do think that Vorbis -q should be considered as well, but if no one other than me and Gecko are interested in it, then it's going out the window.

For the AAC consideration, I was thinking about going with QT (mostly, because I know where I can get an encoder.  Where is the Ahead codec?).  Even though WMA has more hardware support than AAC, I'm not going to touch that one with a 10-ft pole.  Sorry.  Too many unpleasant dealings with it.

One last provision that I failed to mention earlier.  All said codecs must have a free encoder.  I'm interested in the results of the test, but I'm not going to spend money on this curiosity (other than my own time and effort).

Anyone else want AAC QT @ 128, or is there a better setting that you'd like to see?

Are you interested in a Transcoding Listening Test

Reply #13 –
If I can just throw in my two cents worth, seeing as my previous transcoding tests have been mentioned. 

I personally would like to see Wavpack lossy and/or DualStream included, just to see if other people agree with my previous findings. Having said that though, Wavpack lossy/Dualstream can't really compete with mpc for bitrate, so if you were to include them, there will be comeback regarding it being unfair that mpc -standard might be 180 kbits for a sample, where as Wavpack and Dualstream can not produce a file below ~250-270 kbits, and are best considered in the 320kbit range. 

As for a target format, I would be happy with either mp3 or AAC, as they are the most likely to be used, given current hardware support.

As guru mentioned earlier, you need to decide the target audience for this test. If it is portable use, that probably means a low to moderate bitrate. If you have specific portables in mind, you pick mp3 or AAC. If you want to aim for people who make up party music on CDR and play it in their DVD player, mp3 aps, etc.

Whatever the case, I'm happy enough to take part in the test. 

Just as a reference for samples, the beginning of Blue Monday - New Order produces very noticeable artifacts when transcoded from AAC, Musepack, Vorbis and LAME at their maximum settings, into ATRAC. I wouldn't mind betting that you get the same from this going into mp3/aac, but it would be very interesting to find out. The neat thing about this sample, was that for each of the above lossy codecs, it produced different artifacts in the transcode, and some were more annoying than others, so another aspect of a trasncoding test is not just, does it cause artifacts, but what type of artifacts are generated?

Are you interested in a Transcoding Listening Test

Reply #14 –
Quote
As guru mentioned earlier, you need to decide the target audience for this test. If it is portable use, that probably means a low to moderate bitrate. If you have specific portables in mind, you pick mp3 or AAC. If you want to aim for people who make up party music on CDR and play it in their DVD player, mp3 aps, etc.

Whatever the case, I'm happy enough to take part in the test. 

Thank you for being willing to volunteer.  The more people in and listening, the better.

I'll toss WavPack Hybrid @ 320 into the mix, since it's the standard that you judged on (and I figure that most people who would use WavPack would probably do so at that setting).  Anyone wanting to argue against that, now's your chance.

You've made the best argument so far for Lame aps, I think that it should go in.  I knew that mp3-cds are pretty popular, but I had forgotten that many DVD players can play back mp3-cds.  That gives a lot more justification for Lame -aps in my mind.  It's in.

So here's the codecs that I think should be tested:
  • MPC --standard --xlevel
  • MPC --insane --xlevel
  • WavPack Hydrid @ 320
  • Vorbis GT3 -q 6
Target should be:
  • Lame --alt-preset standard
  • AAC QT @ 128
  • Lame --medium?
I put Lame --medium because there doesn't seem to be ANY documentation concerning that.  In fact, I just tried it now so that I could see the file size of it.  Lame threw an error at me.  What exactly is the flag for this medium setting?

Are there any other codec consideration requests?

Are you interested in a Transcoding Listening Test

Reply #15 –
What about a portable preset? That'd probably be a good reason to transcode. Like LAME alt-preset portable instead of medium?
The object of mankind lies in its highest individuals.
One must have chaos in oneself to be able to give birth to a dancing star.

Are you interested in a Transcoding Listening Test

Reply #16 –
Quote
What about a portable preset? That'd probably be a good reason to transcode. Like LAME alt-preset portable instead of medium?

Well, I'd love to have a medium or a portable setting and would be more than happy to include such a setting for a target encoding. The problem is that I know of no such setting.  Lame threw an error at me when I tried the following:
--medium
--preset medium
--alt-preset medium

And according to the documentation listed here, such a feature does not seem to exist.

So I ask again, what setting is there for medium?  Rather, if you were encoding an mp3 for a flash-based MP3 player, what exact lame setting would you use? Alternatively, if there IS a compile that includes the medium flag, could you please tell me so that I can compare file sizes to an -aps -Y encode?

Are you interested in a Transcoding Listening Test

Reply #17 –
--preset medium was introduced by Gabriel Bouvigne with lame 3.93 ; I'm not sure that the 3.90.x branch had this preset. Take lame 3.93.1 if you want to encode music with this new preset

Are you interested in a Transcoding Listening Test

Reply #18 –
John33's modified 3.90.3 compile includes --alt-preset medium, though personally I'd recommend --alt-preset standard -Y.
dev0
"To understand me, you'll have to swallow a world." Or maybe your words.

Are you interested in a Transcoding Listening Test

Reply #19 –
Quote
John33's modified 3.90.3 compile includes --alt-preset medium, though personally I'd recommend --alt-preset standard -Y.
dev0

Just one precision, for people encoding classical music. -Y switch has not here the same effect than on metal... music : few bits are saved. Therefore, --preset standard -Y can't be considered as an universal medium/portable preset.

--preset-standard -Y is better than --medium ; but bitrate is higher.

Are you interested in a Transcoding Listening Test

Reply #20 –
>_< Silly me.  I read the Lame FAQ and just kept the unmodified 3.90.3 compile around, and when it came time to testing the medium, it of course didn't work.

I just ran a quick test on 4 songs.  The modified 3.90.3 is the best of both worlds.  --aps works exactly the same as 3.90.3, and --preset medium works just about the same as 3.93.1

For the purposes of this test, I will use the modified 3.90.3, are there any objections to that?

As for the --preset medium vs --aps -Y case, I've compared the bitrate differences on the same 4 songs.  --preset medium averages at about ~150-160kbps.  --aps -Y weighed in at around ~180-190 kbps.  (These were transcoded from MPC, because I'm on my laptop at work and don't have any CDs with me).  These may not be indicitive of the actual average bitrate since all the songs were off the same album.  Incidentally, Van Halen's Eruption came out to be 208kbps  I think that the Recommended Settings page may have to be updated, because that is nowhere CLOSE to "around ~160".

I know that Van Halen won't necessarily be indicitive of average kbps, but the result makes me lean towards using --preset medium as the target choice for flash-based portables.  I suspect that -aps -Y will yield similar results as -aps

So the target settings will be:
Lame --alt-preset Standard
Lame --preset medium
AAC QT @ 128

One more question.  I don't normally use AAC, so I'm going to need some help on the last one.  Where can I get QuickTime's AAC encoder, or do I have to pay for it?

Are you interested in a Transcoding Listening Test

Reply #21 –
I agree with your findings of alt standard -Y. It doesn't get close to the "advertised" bit rates for most of my music either. Medium's a better bet for a mid 100's type bit rate.

As for Quicktime, you need to pay for it. You need the Pro version to encode at 128kbit. 

Are you interested in a Transcoding Listening Test

Reply #22 –
I love the idea of a transcoding test.

A few things I think would be useful:
1) MP3->MP3
2) MP3->AAC
3) MPC->MP3
4) MPC->AAC

I have a huge (>40 gigabytes) library of mp3s (mostly APS - I usually rip best-of albums and some of my favorite tracks to APX though) and a small but expanding collection of mpcs (which I only encode on a whim - due to non-support on my ipod) which I frequently swap onto my iPod (20gb non-docking with AAC firmware), and it would really be nice to know what kind of quality I can get.

The main thing I'd like to see is to compare MP3 versus MPC converted to lower bitrate mp3s, as that would play a big role in determining my preferred format on my computer. Secondary to that would be the transcoding results to AAC versus MP3 as a target format - as an extension to the above comparison.

And of course, the overall most important thing would be ease of transcoding - in terms of both speed and minimal temporary files (i.e. - decoding straight to the encoder) - it's useless if I can't transcode easily as an intermediate step in transferring to my iPod.

Are you interested in a Transcoding Listening Test

Reply #23 –
Quote
The main thing I'd like to see is to compare MP3 versus MPC converted to lower bitrate mp3s, as that would play a big role in determining my preferred format on my computer. Secondary to that would be the transcoding results to AAC versus MP3 as a target format - as an extension to the above comparison.


Among the perceptual codecs (not WavPack or OptimFrog), MPC originals will give higher quality results. That's because MPC retains higher information at high bitrates as it is and also it is a subband codec which reduces the transcoding artifacts. This feature of MPC is already known on HA. And AAC is very likely to give higher quality result as the target format compared to MP3 because it's a state-of-the-art design transform codec and will always do a better job when the originals are the same. I guess that solves your problems. If you have iPod you don't need MP3 in any of the intermediate steps to begin with...

So I'd safely bet on MPC->AAC transcode to be the highest quality at same bitrates (aps uses too many bits compared to 128 AAC). The only question is how big the difference is... And that's why this test is useful.
The object of mankind lies in its highest individuals.
One must have chaos in oneself to be able to give birth to a dancing star.

Are you interested in a Transcoding Listening Test

Reply #24 –
Quote
As for Quicktime, you need to pay for it. You need the Pro version to encode at 128kbit. 

>_< DOH!!

Ok, do we have any volunteers that would be willing to encode samples into AAC?  If we have no volunteers, I will do the encode in using the Psytel encoder on Rarewares, since that one IS available to the public for free and wasn't all that worse compared to Quicktime.

Also, I/We should determine what samples should be used for testing.  Do you think that the same one's from rjamorim's test would be appropriate, or should we strive for something from all styles?

One last thing to note before you start sending me samples.  I'm going to be out of town and away from my computer next week, so I won't be able hear, compare, or give input.  Should I revive this thread again when I return, or just start a new one?