Lossy Sound Quality

[RMJFlack]

Is there a rule of thumbs as to which of the following will give best sound quality (ignore space factor)?

AAC 256k from iTunes Match
MP3 V0 down-converted form FLAC ripped off original CDS
MP3 256k "

Im assuming that Const 320k is theoretically better but not worth the extra space? The context here is playback thru an iOS device with Harmon Kardon 'phones.
(Id be using FLAC files when played thru my Audio system).

[Postscript added]

I note that the VBR V0 setting produces average bit rates noticeably less than 256k (seem typically to be in range 190 - 230 for many tracks).
The other implied question of course is Apples source for iMatch compared to my CDs (thinking mainly of Classical music here)

[garym]

All three of those are high enough bitrate to be transparent as compared back to the original lossless version. Rare for anyone to be able to ABX a difference between lossy and lossless above, say 192kbs, unless a problem sample. Even lower is fine in many cases. VBR is really the most efficient for lossy as it uses the bitrate it needs for difficult to compress parts of a track and uses less bitrate for easier sections. For myself, I create mp3(lame) VBR at -V2 (which produces an average bitrate of about 192). Of course I have the original lossless FLAC files too, and I use these in my home stereo listening. The lossy mp3 files are for my ipods, iphones, etc. And in fact the -V2 is probably overkill for that purpose. -V4 or -V5 is likely fine.

For yourself, you should do some ABX testing to determine at what level lossy files are NO LONGER transparent. You can do this easily with foobar2000 and its ABX (file comparaator) component. Most people that do this are shocked at the fact they can't tell a good lossy file from the original CD/lossless in a double-blind comparison.

p.s. Apple's source for the iMatch is not likely (actually very UNLIKELY) to be any source different than the 16/44.1 redbook CDs that one would rip from. They don't have special hires "master" files that they create these files from.

[RMJFlack]

Took a very quick run at the ABX thing in FB2000, looks interesting.
It has raised one question for me tho about ABX (which Ive only just discovered.
Am I correct that its purpose is to determine if two tracks can be reliably distinguished.
I assume that to maintain the double blind nature of the trial, the two files are randomly allocated to A & B at the start of the trial. (as opposed to the 1st track is always A).
Presumably the logic is - if they cant, use the smaller file. If they can, use the higher bit rate if space permits?
There is an implied assumption that higher bitrate = better.
That is very plausile where (a) the original source for the two tracks is the same and (b) the same format or enocder is used.
But if one is trying, for example, to compare MP3 V0 from FLAC vs Apple store AAC 256k, theres obviously a lot more in play than just bit rates.
So the issue is then is A "better" than B, which isnt quite the same as is A distinguishable from B.

Or do I have this all wrong? Im new to all this so - Help please!

[garym]

Took a very quick run at the ABX thing in FB2000, looks interesting.
It has raised one question for me tho about ABX (which Ive only just discovered.
Am I correct that its purpose is to determine if two tracks can be reliably distinguished.
I assume that to maintain the double blind nature of the trial, the two files are randomly allocated to A & B at the start of the trial. (as opposed to the 1st track is always A).
Presumably the logic is - if they cant, use the smaller file. If they can, use the higher bit rate if space permits?
There is an implied assumption that higher bitrate = better.
That is very plausile where (a) the original source for the two tracks is the same and (b) the same format or enocder is used.
But if one is trying, for example, to compare MP3 V0 from FLAC vs Apple store AAC 256k, theres obviously a lot more in play than just bit rates.
So the issue is then is A "better" than B, which isnt quite the same as is A distinguishable from B.

Or do I have this all wrong? Im new to all this so - Help please!

Yes, the randomization is done behind the scenes, otherwise not a true double blind test. The entire point of an ABX is whether you can tell the difference between the two tracks. Typically, one of the tracks is lossless and the other is some sort of lossy (mp3). So you are testing tranparency of the lossy codec. But you could also use mp3 vs aac, or mp3 of different bitrates. Doesn't really matter what is at play, the key is whether one can detect any difference. If not, then the codec, the bitrate, etc. is obviously not critical to the listener. ABX doesn't say one is better, it merely tests whether the listener can detect whether the two tracks are *different*. And it turns out that it is quite difficult with normal music to detect differences between lossless originals and lossy versions, so long as the lossy version is good enough (which for many means a bitrate of about 192kbs or more). More here:

http://en.wikipedia.org/wiki/ABX_test

[garym]

Yes, the randomization is done behind the scenes, otherwise not a true double blind test. The entire point of an ABX is whether you can tell the difference between the two tracks. Typically, one of the tracks is lossless and the other is some sort of lossy (mp3). So you are testing tranparency of the lossy codec. But you could also use mp3 vs aac, or mp3 of different bitrates. Doesn't really matter what is at play, the key is whether one can detect any difference. If not, then the codec, the bitrate, etc. is obviously not critical to the listener. ABX doesn't say one is better, it merely tests whether the listener can detect whether the two tracks are *different*. And it turns out that it is quite difficult with normal music to detect differences between lossless originals and lossy versions, so long as the lossy version is good enough (which for many means a bitrate of about 192kbs or more). More here:

http://en.wikipedia.org/wiki/ABX_test

And in spite of the fact that I can't ABX any difference between an mp3 192kbps track and lossless, I still rip all my CDs to lossless (FLAC). I use these at home and then make mp3 (lame) lossy copies at -V2 for use in portables. Having the FLAC files is good because one can always convert these to any other format without loss. One can't do this with a lossy file (that is, going from mp3 to mp3, or mp3 to m4a, is a lossy process and not a good idea).

[RMJFlack]

Thanks that confirms what I thought.
Is there anything to help with the "which is better" question?
And yes as I suspected youre right about the last question. My one quick test, I was 3/4 correct between FLAC and mp3 V0 played thru my desktop speakers (an opera track), but the difference was very subtle.

[RMJFlack]

Yes the basic logic of all this is pretty clear.
My main questions at the moment are: Apple 256k from iMatch vs v0 mp3 from my FLAC for iOS versions of the good stuff; and is the quality / space / effort worthwhile to make V2 or even lower copies of some secondary tracks to get more on my iPhone?

[garym]

Thanks that confirms what I thought.
Is there anything to help with the "which is better" question?
And yes as I suspected youre right about the last question. My one quick test, I was 3/4 correct between FLAC and mp3 V0 played thru my desktop speakers (an opera track), but the difference was very subtle.

be a bit careful. When looking at ABX results, you're seeing the 'p-values' of the statistical test of rejecting the "null" hypothesis (no difference). So a p-value of say, 0.75, doesn't mean you got 75% correct, it means that you were way, way off from being able to tell a difference. A p-value of say, 0.05, means you are able to reject the null of "no difference" and thus you *can* tell a difference between the tracks. An easier thing to look at is at the end where it tells you the "hit rate" (how many correct choices you had out of the total attempts). So typically one would do 20 trials. If your hit rate is 16/20, then you were able to correctly identify the track 80% of the time (which is good). But if you are 10/20, this is 50% and no better than simply guessing.

Regarding true quality and preference, only one's ears and brain can decide this. But the ABX test showing that you can't tell the difference is good evidence that the quality is perceptually the same no matter what (to you and your ears/brain). Regarding what to keep, I'd prefer a mp3 -V0 version from my own FLAC rip (where I KNOW the rip was done properly and I may even have an accuraterip match) to a itunes match 256 version. Nothing wrong with itunes match, and I may even do that for some of the stray mp3/aac files I have from years ago. But for my own CD rips, I prefer my own FLACs, then create lossy from those. And for just portable use (when you have the FLACs), I'd say -V4 or -V5 is more than enough for your iphone and headphone listening.

[RMJFlack]

Yes it was 3 right out of 4 tries. Obviously very small sample.
Home-made V0 vs iMatch - what you say makes sense and is where I was headed.
And the interesting thing will be to see how low I can go with my headphones or desktop speakers.

[garym]

Yes it was 3 right out of 4 tries. Obviously very small sample.
Home-made V0 vs iMatch - what you say makes sense and is where I was headed.
And the interesting thing will be to see how low I can go with my headphones or desktop speakers.

yep, 4 trials really way too low. You should try an ABX with your headphones and do at least 20 trials. Should have replaygain turned on too during the ABX test, as very slight differences in volume can make tracks identifiable.

[RMJFlack]

Hope this isnt a silly question - but when comparing lossy files in different formats / bitrates, could the player affect one's preference? Im talking mid range equipment here (ie iPad + phones [Harmon Kardon] vs LMS + NAD / B&W Towers

[garym]

Hope this isnt a silly question - but when comparing lossy files in different formats / bitrates, could the player affect one's preference? Im talking mid range equipment here (ie iPad + phones [Harmon Kardon] vs LMS + NAD / B&W Towers

Only to the extent a player can't actually play the type file (that is, some players can't handle m4a/AAC). For ipad, it will handle m4a, mp3 just fine.

[Porcus]

General considerations first:

- There is no guarantee that the iTunes Match version is actually the same version as what is ripped from a CD. Different masterings do sound different.
- If one worries about whether 200+ bitrates are sufficient, then just go lossless. Lossless has other advantages (e.g. can be verified to be a first-generation error-free rip).
- If you have a lossless archive, and want to transcode to a portable player, then why not try 96 or 128, and just overwrite with a better transcoding those albums which sound annoying?

Should have replaygain turned on too during the ABX test, as very slight differences in volume can make tracks identifiable.

That means: use the RG feature when using fb2k's ABX component. You do not have to set fb2k's playback settings nor pre-scan tracks, as the ABXer will scan.