Hmm... okay, thanks for pointing out those links, will for sure study it.
I know that lossless doesn't make sense for portable use, even if I have 250$ portable headphones - the small digital audio interface integrated into a palmtop can hardly be concidered audiophile...
But, the problem is with the recompression. I have, like, 250GB of lossless music on my HDD. It would take days if I were to recompress it all to OGG and keep both versions of the same music files (that would be another, what, 150-200GB ? ).
Now I can just get a 16GB SDHC card for my PPC and without wasting much time, drag and drop the albums I want to listen for this week and I'm done.
That's why I want to settle for one universal standard. And I do hear the diferrence between lossless and not lossless. (well acctually OGG is extremally good at the higher settings, but the filesize is almost the same as flac/wavpack so I just prefer to keep the untouched file).

Thanks!
Have a great day!

Hardly! OGG Q2 is roughly 6% of the original lossless size, you would be looking at an additional 21GB (250GB * 6%-lossy-encode / 70%-FLAC-compression). And this process can be completely automated with FooBar2000.


I'll take your word for it. If you want higher bitrates, then I am not going to argue with you about it. But I highly recommend you take 4 of your favorite songs which constitute different styles/genres of music and ABX them against... LAME(3.96b6) -V2 MP3 which is ~256kbps, or ~16% original size (again looking at only 57GB)... if you hear a difference then go with FLAC tracks. If you can't tell the difference then just go with FLAC images, throw them to DVDs when you have enough of them, and call it a night.

I do use foobar2000 extensively, also for recompression, so I know all about the automation and it's great ability to make one's life a lot easier

I did similar tests to those described by you some time ago, but I will repeat them and post my results. But I can absolutly hear the diferrence between MP3 at 320kbps and lossless file. Maybe not on all kinds of music, but the ones that I usually listen to have lots of details that are lost. Heck, I can even spot the diferrence between lossless 16-bit and lossless 24-bit file. I did "blind" tests, a friend converted a 24-bit sample of my choice to 16-bit WAVE, to MP3 at 320kbps and to OGG at highest lossy 24-bit setting (which was somewhere at about 450-500kbps VBR) - only when he played the OGG file I wasn't able to spot the diferrence.
All it takes is a good example, good headphones with a decent audio interface. You don't even have to be an audiophile or have incredible hearing.

I listen mostly to jazz, classical and progressive rock. I also work in the sound design industry where the detail really counts.

Ah you have many ups on me. I have a hole in one ear, at best cheap-shotty headphones, and only work only in the digital-side of a DJ business. But if you can hear that... that's fantastic! I wish I could hear that well. There are plenty here that would value your input to improve their lossy codecs.

I do personally not believe that you are able to do this, so please proove me wrong by posting some ABX tests here...

If you are talking about the 16-bit vs 24-bit almost anyone with decent hearing can get the difference. I bet you yourself can. You just need appropriate samples (very subtle and quiet ones) and a decent 24-bit external (or at least one that can filter lot's of interferences) sound card. Good headphones are welcome too.
I do a lot of field recordings myself and you can spot the difference listening to the reverb tail. Or just listen to a recording of a forest after midnight. In 16-bit's you will loose a lot of details.

As for the 320kbps MP3's vs Lossless files - if your hearing allows you to hear above 16kHz - you can feel it too (especially on jazzy percussion samples - crash, china - all the hardware, but hi-hat especially and on the reverb tails one can hear a bit of "crystaling" effect).

I will do the ABX tests, as I'm very curious myself.
And since the big factor with this is very often a psychological thing (the subconscious mind can tell itself that something is better and it will actually feel better - in this case, the lossless file) it's probable, that to some point, I'm experiencing this myself.
This reminds me about an anecdote about audiophiles. There was a blind test for cables performed, to which about 10 audiophiles where invited. The same sample was played through a great, pricey equipment using cheep RCA cables for like 5USD and through really expensive cables for 2000$ a meter.
Those audio freaks where able to distinguish, and spotted the difference, but... no one was able to tell which one is the pricey one. It sounded different, but no one could tell if the 2000 dollar one is better then the 5 dollar one.

And again, the 24bit vs 16bit - the difference is there, you just have to know what to look for - the "silence" or the "near silence" parts, which sound a lot deeper in 24-bit.

How often do you listen at a volume where you can distinguish this difference when the data is near full-scale regularly within a track of music?

I request that this thread be split.

Indeed, we got a little off the topic. And I agree with you, the diferrence is often not that noticable. Well, I'm not an audiophile, I often listen to lossy audio. As I don't wan't to pointlessly argue, I completly agree with you that 24-bit is usualy an overkill and so is lossless audio in compare with high quality VBR-compressed vorbis file.

Moderator: Sorry for this OT !

Going back to topic, I think I'll stick to multifile flac, as it is most supported and has good compression to size ratio, and doesn't require a fast processor (works flawlessly even on downclocked to 150MHz PPC).

Have a great day!

Please just be sure to post samples (30 seconds or less) and your ABX results. Our TOS specifically states that you cannot make such claims without proof.

The chances of noticing any difference between 24-bit and a properly dithered 16-bit source at normal listening levels are practically zero, even on good equipment.

it sounds like there are other variables in the process...

I think you are hearing a slight difference in volume which can happen when converting to different formats, and thinking the slightly higher volume sounds better.

Not at all, as I also know of "louder is better" blind test (people who hear two identical samples, of which one of them is 1dB louder will say that "it was better", and it's an indefinate loop - one would be able to make a "better" sample by boosting the volume by +1dB each test). I was talking more about the close-to-silent phrases that just have better depth. But, acctually danbee maybe right - it might aswell be a matter of good dithering (maybe I was comparing it on a "badly" dithered sample).

Again, I will do the tests in the near future (I'm very busy lately, I have about 65 hours a week of work) and post them here.

As for this:

I wasn't even aware of the existance of ABX for foobar2000 until yesterday, that's why I made such stupid claims I should stab myself in the face for not reading the TOS before posting, too.

I do not know, for fact, that I'll be able to hear it nowadays . But I'll try, and, well, HOPE that I do, so I won't make a complete ass of myself.

Sorry again, and have a wonderful day!
NIXin

Hi NIXin.

I just want to say that you're really being a good sport about all of this.

Regarding the test @320kbits, make sure you specify what encoder you're using. I think most of us here recommend that you use either 3.97 or 3.98 of Lame and not use any extra settings other than to specify the bitrate. It might be fun to try Blade as well just for giggles.

And one should never make claims that two items are identical, because ABX tests can NEVER prove such a thing (these tests can only prove that they are different).

I have seen several posts around here saying things like "mp3 sounds the same as lossless and I have the ABX test results to prove it" that deserve the same form of punishment.

Isn't that the point of ABX though? To prove that two files either sound identical or different? One could test 320kbps Lame 3.98b6 vs WAV and fail the ABX test. That would prove that one could not effectively distinguish between the two thus making the files sound alike to the listener. Failing a ABX test would show that the files are identical to the person taking the test. Remember that all ABX tests are subjective to the person taking them. Me failing a ABX test comparing FLAC and iTunes AAC at 128kbps VBR just tells me that iTunes AAC at 128kbps VBR suites my lossy encoding needs.

No no!! That's not allowed because people can fail a test on purpose. So this doesn't prove anything.

Imagine that you have mp3's of two different songs. In your theory I can easily prove that they sound the same by failing an ABX test between the two 16/16 times. That's not science

The hypothesis is: "A and B sound the same" and you can only reject this hypothesis, not prove it.

[edit] I would like to add that you can use ABX as a tool for yourself to check whether a certain codec/setting is transparent to you. You yourself knows best if you did the test in an honest way. But again, you cannot use this as a scientific proof. [/edit]

Well it would make more sense if you were to fail eight out of sixteen.

It's also possible that you could differentiate between a pair of encodings that you could not earlier once you learned what to listen for.

OK, a have done some preliminary tests. Man, it was hard ! I didn't hear any difference on 5 samples. The 6th, which was a 10 second fragment of Imaginary Day by Pat Metheny, was my savior Finally, heard some difference (the bells get a bit "crystallized" and the spectrum is smaller). The difference isn't that obvious aether (hence my score 8/10 in two runs - it's difficult to concentrate on the details). I did the first test on my speakers, the other one on headphones. Nether of them can be called high-end, I'll do the ABX again when I'll be in my studio with far superior setup.
My home setup: NAD 3020, Tannoy M2, Alesis IO|14 and the headphones I was using was KOSS PORTA PRO. I also have here, at home, Ultimate Ears Super.Fi 5 Pro, but their "supported" range limits to <16kHz, so I guess the test wouldn't be fair. ;-)

I might have score better on better equipment.

A file: E:\TEMP\ABX\6-frag.wav
B file: E:\TEMP\ABX\6-frag.mp3.wav

Start position 00:00.0, end position 00:10.0

10:39:10 1/1 p=50.0%
10:39:31 2/2 p=25.0%
10:40:21 3/3 p=12.5%
10:41:47 3/4 p=31.2%
10:42:20 4/5 p=18.8%
10:42:57 5/6 p=10.9%
10:43:16 6/7 p=6.2%
10:43:33 6/8 p=14.5%
10:44:15 7/9 p=9.0%
10:45:17 8/10 p=5.5%
10:48:33 reset

10:51:15 1/1 p=50.0%
10:52:57 2/2 p=25.0%
10:53:43 3/3 p=12.5%
10:54:17 4/4 p=6.2%
10:54:51 5/5 p=3.1%
10:55:18 5/6 p=10.9%
10:56:05 6/7 p=6.2%
10:57:08 7/8 p=3.5%
10:58:03 8/9 p=2.0%
10:58:45 8/10 p=5.5%

Have a nice day.

PS
The samples I was using can be downloaded here:
http://aurinko.nixion.net/~nixin/public/ABX-Test1/

I compressed with LAME 0.97 (-b 320)

I must admit that MP3 is far superior when comparing it to what it was couple of years ago - that's probably why I remembered there was a easy noticeable difference. It's not that noticeable at all.

Have a great day!

/edit/ I should probably add, that I was tested on an audiometer not that long ago, I'm able to hear up to 18kHz w/o specially listening for something, and up to 22kHz when the source is loud/very loud. That might have something to do with the results too, because of the lowpass filter - according to LAME the transition band was at 20094 - 20627 Hz. That's probably why the sound may have been a little duller for me.

My comment was not about sense

I guess there's no sense at all to failing the ABX on purpose.

If you got a score of 0/16 (or 16/16 all fails) and suggested to someone that it was because you couldn't tell the difference you'd be sussed out as a liar (and a dumb one at that).

Of course you may have been honest but extremely unlucky.

Congratulations: you do hear differences between Lame 3.97 CBR 320 and the original, and you did so with your 6th track you tried. That's very good.
So you are a valuable member here as you do have fine hearing.

As for your initial question:
You are able to hear differences when giving all your best within a situation that has nothing to do with practical listening even when listening carefully, but it looks like the differences were subtle and they are audible only on rare occasion.
That's the world of mp3 (and other formats), and it looks like it's not a real problem to you when used with your palm.

You may want to consider variants:
The current 3.98 development (it's in beta state) is very promising with respect to improving quality even further. Use 3.98b5 if you want to give it a try (with the currently published 3.98b6 it's not totally clear whether the Lame devs wanted to have it published at the current state - it's not the Lame devs who do the publishing).
You can save some disc space by allowing for a lower bitrate than 320 kbps. IMO using --abr 270 yields the same quality in a practical sense as -b 320 while saving ~15% of disc space. It's not necessarily 270, and you can try a lower value. Most people here prefer a VBR method, and with 3.98's VBR has no issue any more according to my experience. So you may want to try for instance -V1 to get at a bitrate of ~225 kbps on average. As quality is paramount to you I wouldn't try other settings however yielding even a smaller filesize though usually quality will be fine as well.

In case you don't like mp3 etc. despite (or because?) of your test you may want to give lossyWAV a chance.
lossyWAV makes those least signifant bits of the wave samples to zero in a way that is done safely - safely according to the principles of lossyWAV. Using an appropriate lossless codec like FLAC on the lossyWAV result provides for an average bitrate of ~350 kbps when using LossyWAV quality level -3. All our listening experience so far shows up transparency even for this lowest quality setting.
So if your DAP is able to play FLAC or another appropriate lossless codec, you can go this way.
Look at the lossyWAV thread in case you're interested.

Please come on, don't be a bad "loser" .
He provided the requested evidence that he can hear a difference. AFAIK there are no restrictions to the playback conditions, as long as they can be considered "normal" (no unusually high monitoring levels to reveal low-level artifacts e.g.)
If NIXin is able to hear up to 22 kHz, that could as well be an explanation for his ability to hear differences between 24 and 16 bit audio, when noise-shaped dither has been used. Some dither versions put lots of energy in the higher frequencies which are presumed inaudible for most humans, but perhaps not for him. Flat or moderately shaped dither might be a better option in that case.

Kees de Visser

Not to cast any aspersions, because I certainly believe that NIXin is hearing what he reports, but I think that 8 out of ten should probably be considered borderline confirmation, with a 5.5% probability of random selection. It helps that he achieved this twice, but in future he might want to increase the number of trials.

??? This was an introduction to the practical consequences of NIXin's test, and sure the abx situation has not much in common with practical listening.

This doesn't reduce the value of NIXin's abx test, and I gave him congratulations for his success.

I just wanted to clarify what i meant with the above...

Of course i believe that it is possible to ABX 320KB MP3 against lossless on various MP3 problem samples and if having good hearing, however, what i stated to not believe where solely that someone should be able to tell the difference on all the music he normally listened too and that there where lots of details that would get lost in general with 320kb MP3. Other than that, then i also didn't believe that somebody could abx 16 bit lossless against 24bit lossless on normal music playback.

Other than that clarification, then i just wanted to say thanks for the abx report, NIXin - and that i also think that you are being a good sport about this whole thing