Topic: Is There A Way Of Calculating A Ratio of Quality Lost (FLAC vs mp3C)?

[DMBprez]

Anyone have any idea what kind of quality you actually lose when converting from FLAC to 320kbps?

Just curious.


Thanks.

[acidjack]

Anyone have any idea what kind of quality you actually lose when converting from FLAC to 320kbps?

Just curious.


Thanks.

Good question. I have often been curious if there has been any empirical analysis of FLAC to maximum-quality VBR files (AAC or FLAC). I assume, perhaps wrong, that since archive.org uses VBR, it must be better, since it theoretically provides "smarter" compression.

[Scooter123]

Its imperceptable to my ears--@320kbs, even with the most critical music and my best speakers, Vandersteins + Vanderstein Sub-Woofer. 

[rastasean]

well maybe we need to ask what quality is lost going from wave to flac level 6 so we can understand what frequency range is lost going from flac level 6 to mp3 320kbps.


comparing two of the same audio files is probably nearly impossible for most people but what about comparing a raw photograph (as in the format, raw) to a very, very high quality jpeg. would we be able to distinguish the difference?

[Will_S]

well maybe we need to ask what quality is lost going from wave to flac level 6

None at all - that's the whole point of lossless compression, the original data is recreated 100% upon decompression, it is just stored in a more efficient format.

so we can understand what frequency range is lost going from flac level 6 to mp3 320kbps.

It's not a simple matter of a particular frequency range being lost.  What information gets lost is very much a function of the original source material and the particular encoder used.  320 kbps mp3 is not a sufficient description to know exactly what encoding steps were performed, different encoders make different "choices" about what information is ok to lose, after first stripping out "redundant" information that can be recreated perfectly by the decoder.

comparing two of the same audio files is probably nearly impossible for most people but what about comparing a raw photograph (as in the format, raw) to a very, very high quality jpeg. would we be able to distinguish the difference?

As you are probably aware, jpeg is a lossy compression scheme (like mp3).  An analogy to flac would be making a zip file of the raw image - the information is 100% preserved.  Beyond that I'm not sure how the comparison is relevant - a highly compressed mp3 is obviously audibly inferior to the original wav file, while a higher bit rate mp3 made with a better encoder may be audibly indistinguishable.  But analyzing it on a bit for bit level you'll be able to detect a difference, just like a pixel for pixel comparison of a raw and jpeg image.  There's no reason to expect a certain % reduction in file size via jpeg compression and the same % reduction via mp3 compression to have the same impact on how close the compressed file comes to reproducing the original, the compression algortithms and the information needing to be preserved are just too different.

[Will_S]

Anyone have any idea what kind of quality you actually lose when converting from FLAC to 320kbps?

How much "quality" is lost is entirely subjective.

How much "information" is lost, you might get a first approximation by comparing the bit rate of the resultant flac file (using the most complete, slowest to excute compression) to 320 kbps.  I.e. if the flac file had a final size that equated to 400 kbps, you might say the mp3 lost 20% of the nonredundant information.  Of course, that assumes that the FLAC process is 100% efficient at only leaving important information behind, and that the mp3 process is 100% efficient at removing all redundant information before it starts removing nonredundant information.  Neither is likely to be entirely true in practice.  But you will find that wav files containing less information (i.e. low peaks, so they don't use the full bit depth, or not much stereo separation, so there is a lot of redundancy between left and right that is more efficiently represented with a mid signal and a side signal that can be fully described with fewer bits since it has little dynamic range) compress into smaller flacs than ones with lots of dynamic range and lots of stereo information.

[DMBprez]

Alright, thanks for the information guys. 


Now: I want to throw a situation out there, kind of curious of the reactions around here:

You upload a tape online (etree, whatever).  Are you okay with people distributing in FLAC?  I feel like Archive's atmosphere has changed the entire trading community BIG TIME.  Curious on your guys takes on it.

I agree with Scott in that there is literally no audible difference when listening  to FLACs and 320 mp3.  What do you guys feel?

[OFOTD]

You upload a tape online (etree, whatever).  Are you okay with people distributing in FLAC?

I agree with Scott in that there is literally no audible difference when listening  to FLACs and 320 mp3.  What do you guys feel?

I think you'll find that FLAC is absolutely the preferred distribution container.

Audible differences?  Absolutely no question about it.        If you're comparing mp3's and FLAC's on cheap computer speakers or cheap ipod headphones then possibly and probably no.   But throw on a a good set of headphones or a decent playback system and you will without doubt be able to tell the difference.

[Will_S]

Audible differences?  Absolutely no question about it.        If you're comparing mp3's and FLAC's on cheap computer speakers or cheap ipod headphones then possibly and probably no.   But throw on a a good set of headphones or a decent playback system and you will without doubt be able to tell the difference.

This is definitely true for the standard 128 kbps dreck that marks most people's experience of MP3s.  But can you honestly say you are confident you can distinguish a well-encoded 320kbps file from the original source in a blind test?

[rhinowing]

forget mp3, what about a V0

[OFOTD]

Audible differences?  Absolutely no question about it.        If you're comparing mp3's and FLAC's on cheap computer speakers or cheap ipod headphones then possibly and probably no.   But throw on a a good set of headphones or a decent playback system and you will without doubt be able to tell the difference.

This is definitely true for the standard 128 kbps dreck that marks most people's experience of MP3s.  But can you honestly say you are confident you can distinguish a well-encoded 320kbps file from the original source in a blind test?

Yes I feel confident that I could on my own playback.  Without doubt.   Now again on cheap PC speakers or ipod headphones possibly not but otherwise I believe I can.

[boojum]

Just a short story about people who "know" they can distinguish MP3's from WAV files in a blind test.  A fellow over at EAC support had a Golden Ears friend who could definitely and beyond a doubt distinguish MP3's from WAV's and gladly accepted the opportunity to prove it.  So the tester played a series of recorded tracks and asked could he tell which was which. Of course!  And in no uncertain terms he proceeded to say "A" was this and "B" was that and so on.   When the test was over and the testee had finished with his opinions the tester revealed that all the tracks were MP3's, at 128kBps.  The testee never again spoke to the tester.

Moral of the story: do not be too sure until you have actually done the test,  "One test is worth a thousand opinions."

And, you may have missed this post above:


"Its imperceptable to my ears--@320kbs, even with the most critical music and my best speakers, Vandersteins + Vanderstein Sub-Woofer. "

[OFOTD]

So because two people couldn't hear the difference that means the rest of the world can't either?

boojum are you a skeptic about everything unless it's in a book somewhere?

[DMBprez]

Just a short story about people who "know" they can distinguish MP3's from WAV files in a blind test.  A fellow over at EAC support had a Golden Ears friend who could definitely and beyond a doubt distinguish MP3's from WAV's and gladly accepted the opportunity to prove it.  So the tester played a series of recorded tracks and asked could he tell which was which. Of course!  And in no uncertain terms he proceeded to say "A" was this and "B" was that and so on.   When the test was over and the testee had finished with his opinions the tester revealed that all the tracks were MP3's, at 128kBps.  The testee never again spoke to the tester.

Moral of the story: do not be too sure until you have actually done the test,  "One test is worth a thousand opinions."

And, you may have missed this post above:


"Its imperceptable to my ears--@320kbs, even with the most critical music and my best speakers, Vandersteins + Vanderstein Sub-Woofer. "

Hahahah, awesome.

[DMBprez]

So because two people couldn't hear the difference that means the rest of the world can't either?

boojum are you a skeptic about everything unless it's in a book somewhere?

I'm in no means doubting you, because I am most certainly not an audiophile.  But I am curious as to what differences there are?  As we talked about, when you convert the file, it destroys "unnecessary," data, so I'm curious what that does to the sound.


Again, I don't mean this in an attacking way.  I'm really just curious.

[OFOTD]

I'm in no means doubting you, because I am most certainly not an audiophile.  But I am curious as to what differences there are?  As we talked about, when you convert the file, it destroys "unnecessary," data, so I'm curious what that does to the sound.


Again, I don't mean this in an attacking way.  I'm really just curious.

Sharp fatiguing highs, flat lows, a perceptible amount of digital fuzz over prolonged listening and most certainly a smaller dynamic range.   As has been stated for years, load up an mp3 and FLAC from an original source and compare them in your audio editor of choice.  Use the spectrum analyzer or similar feature.   

Just because you or boojum may not hear the differences doesn't mean that they are not there.   

[Will_S]

So because two people couldn't hear the difference that means the rest of the world can't either?

boojum are you a skeptic about everything unless it's in a book somewhere?

I'm in no means doubting you, because I am most certainly not an audiophile.  But I am curious as to what differences there are?  As we talked about, when you convert the file, it destroys "unnecessary," data, so I'm curious what that does to the sound.


Again, I don't mean this in an attacking way.  I'm really just curious.

I don't think anyone said the mp3 process destroys unnecessary data.  Rather, a lot of the data it takes out is redundant data that can be recreated perfectly, same as FLAC.  But to get the files even smaller (and lossless compression typically does not get down to 320 kbps) nonredundant information needs to get taken out as well.  How audible this is is debatable, and obviously how audible it will be depends on how much information is taken out and how clever the filters are about what information they take away.

That said, it's far more than just two people who have made very confident statements they can hear differences only to completely fail to do so in blind testing.  So, I always take "sighted" reports of audible differences with a grain of salt.  Expectation bias is a strong thing.  Especially when one's mind has been poisoned against the very idea of lossy compression by the first generation of 128 kbps mp3s, which I'd agree have the characteristics OFOTD names.  But a well encoded 320 kbps mp3 is a different beast entirely.  Some folks may hear a difference, but I'm unaware of anyone who has proven the ability to do so in a blind test.  Maybe it's happened, if so I'd love to hear about it.

[SmokinJoe]

I'll ask the stupid question....

My general method for making MP3s is a command line "lame --preset extreme" as shown below.  I've wondered... does this line below mean it's discarding the information above 19,xxx hertz?  If so... that's fine because my 46 year old ears can't hear that on any system I have (typically earbuds on an iRiver). 

The question in my mind is this... does it imply that it keeps what's below that?

G:\tunes\ice_storm>lame --preset extreme ice_storm_2008-12-12_sample1.wav
LAME 3.97 (beta 2, Jan 24 2006) 32bits (http://www.mp3dev.org/)
Using polyphase lowpass filter, transition band: 19383 Hz - 19916 Hz
Encoding ice_storm_2008-12-12_sample1.wav
      to ice_storm_2008-12-12_sample1.wav.mp3
Encoding as 44.1 kHz VBR(q=0) j-stereo MPEG-1 Layer III (ca. 5.7x) qval=3
    Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA
   423/423   (100%)|    0:01/    0:01|    0:02/    0:02|   8.9472x|    0:00
 32 [  1] *
 40 [  0]
 48 [  0]
 56 [  0]
 64 [  0]
 80 [  0]
 96 [  0]
112 [  0]
128 [  0]
160 [  1] %
192 [  0]
224 [ 73] %%%%%%%%%%%%%%%%%%%%%%%%%%%
256 [155] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
320 [193] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-------------------------------------------------------------------------------
   kbps        LR    MS  %     long switch short %
  278.9       99.8   0.2        90.3   5.4   4.3
Writing LAME Tag...done
ReplayGain: +0.3dB

(by the way... the above sample is sound from a winter time ice storm... lots of ice/rain drops.... hence the high frequency side of the scale.  Regular music tends to bunch up a little lower, maybe 224 and 256, but those 19,xxx limits seem to be the same).

[Gutbucket]

The question in my mind is this... does it imply that it keeps what's below that?

Not neccessarily..

It seems the encoder you are using low passes the signal first before doing any other processing, and that low pass filter has a transition band between 19383 - 19916Hz.
The transition band is the sloped part of the frequency range beween the corner frequency (which is where the filter starts working) and the stop band (which is were the cut-off is essentially complete).

That part is pretty simple in audio terms.

The complicated part is all the stuff that happens after that.  MP3 and other lossy compression formats use perceptual encoding which means that the information they choose to discard after that simple low pass filter are things that the codec creators have decided humans don't hear well.  There lies the fuzzy art of perceptual encoding.  It does more than this but as an example: if there is a quiet sound that happens simultaneously with a loud sound, then the loud sound will tend to mask the quiet sound, so the quiet sound is thrown away.. hopefully without a human noticing.  MP3 is specifically tailored to the way humans hear.  It may not work as well for other animals.

So it does not keep everything below the low pass frquency, but is clever about what it decides to discard.

I'm sure there are much better descriptions of how the process works, but at least that should give you a basic idea of part of what is going on.

[OFOTD]

The tough thing about the direction of this kind of thread is that it ends up like cable threads.  I hear it, no I don't hear it, maybe I hear it..........  blind test, multiple file, double blind......... this testing method, that testing method,......  this book says, that book says.....

To the OP do you hear a difference?  What do you listen back on?   How long do you listen for at a time usually?

Then the question becomes can you tell a difference between different quality playback devices. 

MP3 at whatever bitrate may be perfect for folks others not so much.  Everyone has a different perception of not only what they hear but how much they do or don't care for the intricacies of their source material.

 

[boojum]

So because two people couldn't hear the difference that means the rest of the world can't either?

boojum are you a skeptic about everything unless it's in a book somewhere?

Just show me the evidence.  One test is worth a thousand opinions.  I have been down these roads and heard this stuff way longer than you.  That is why I am a skeptic.   ;o)  I have for 55 years heard the stories of Golden Ears and super gear again and again.  It is not impossible, but I am a skeptic.  Listening to your set one time with FLAC/WAV files and the MP3's is not a test.  When someone else plays unknown files for you, on your set if you wish, in an order neither of you know which is which in, that is a test.  And yes, I believe published data way more than the anecdotal BS that floats around unsubstantiated. 

Prove me wrong.  Take a double blind test with a witness and let us see the results.  That would be fair, wouldn't it?    ;o)  You see, it is just that I have heard this stuff for decades but rarely have I seen it pass the rigorous test.  You do not have to believe what I believe but I have the right to believe it and support it.  And if you have a reason that anecdotal evidence is better than rigorous testing, tell us about it.

[fmaderjr]

Use the spectrum analyzer or similar feature.   

Just because you or boojum may not hear the differences doesn't mean that they are not there.   

I tend to agree with Boojum. I don't save files in 320 mp3, but I sometimes make a Hi-MD recording in Hi-SP mode to avoid a disc change (which I think uses as much data a 256 mp3) and I absolutely hear no difference with great headphones. The argument wasn't about whether there was a difference-it was about whether you could hear a difference. There's a big difference between being able to see a difference in a spectrum analyzer and being able to hear it.

I realize OFOTD would be incensed by this because he once lambasted me for saying I preferred to record at 24/44.1 with a Korg MR-1 instead of using up disc space with DSD, but this is a hobby and we should all do what we are comfortable with.

[OFOTD]

but this is a hobby and we should all do what we are comfortable with.

You are totally right with that statement.

For someone though to tell another person what they do or do not hear is just ridiculous.   Just because boojum cant hear the difference doesn't 1. mean it's not there and 2. that person a or b also can't.   

[Will_S]

Agreed.  So long as people think they can hear a difference, they'll experience more enjoyment listening to the one they think sounds better, regardless of how they would perform in a blind test.  More power to them.

[boojum]

but this is a hobby and we should all do what we are comfortable with.

You are totally right with that statement.

For someone though to tell another person what they do or do not hear is just ridiculous.   Just because boojum cant hear the difference doesn't 1. mean it's not there and 2. that person a or b also can't.   

Wrong, again.   I never said what I could or could not hear.  I merely pointed out the results of what others have tested.  The difference between anecdotal and test.

Not all MP3 files are the same.  This is not WAV or FLAC.  Older MP3 files at 128 were crap.  The recent releases of LAME, above 3.8, are pretty bullet proof, as has been shown by people on this board.  I really suggest you take a good test of the recent LAME release, 3.98, at the higher rates, I like V2, and see how it sounds before you say it does not work and that a difference can be heard.  This is just an idea.  You need not follow it.

[OFOTD]

before you say it does not work and that a difference can be heard.  This is just an idea.  You need not follow it.

Again who are you to tell a person what they can and cannot hear? 

I am 100% comfortable with my ears and the playback I use to determine what I do and do not hear.   In the end no matter what you may think mp3's be them 128 or 320 ARE lossy compression schemes even with the newer schemes.   This I believe has been proven earlier in the post quite well already.   

As I have previously stated with how I hear the difference in mp3's for me a test is pretty easy to pick out the inferior source.  Bass response is a quick way to tell. 

So what you're trying to prove boojum is what you hear and what you read and how it relates to what others may or may not hear.   You've used a similar line of attack in the cable threads as you believe them to be snake oil.  Agree to disagree because I can't make you hear a difference in cables but they are there.  Same with mp3's vs. FLAC.  I can't make you hear anything you seem predisposed to not hearing anyways.   So a debate with you on this subject is really pointless because you are not either able or willing to to see another side and rely solely on a double blind test method to be the only way to tell.   That's your limitation in your thinking not mine.

We could have 100 people here say 'Yes I hear a difference' but that wouldn't matter to you as you are firmly in the camp that can't tell a difference.   What can someone who clearly hears something do to convince you that they can hear it?   

Additionally who is to say that in the future that a lossy compression scheme couldn't rival a lossless scheme.  I won't go that far but as of May 6th, 2010 I believe that I can tell a difference.  YMMV though.

[kirk97132]

So what you're trying to prove boojum is what you hear and what you read and how it relates to what others may or may not hear.
 You've used a similar line of attack in the cable threads as you believe them to be snake oil. 

QFT.  (unfortunately)   

"Just because you can't see the wind does that mean it's not there"?

The OP's wording of " Anyone have any idea what kind of quality you actually lose when converting from FLAC to 320kbps?"  Leaves the door open for a lot of ambiguity.  What does quality actually mean?  Possibly a better way to word this would be to replace quality with information.   But to ask what quality is like asking what is good......or bad for that matter.   Another question could also be what are the average losses in hearing at ten year intervals?  And across what frequencies do they occur?  Then add in the variable of what circumstances accelerate hearing degeneration IE: loud work environments like production shop, Military service, loud music exposure etc.  So it could be that the older a person is the less accurate the information they hear.  The less accurate the information the more likely to incur a biased perception.  Just because you believe it does not make it right.

[setboy]

You upload a tape online (etree, whatever).  Are you okay with people distributing in FLAC? 

Why would i not be ok with that?

[DMBprez]

You upload a tape online (etree, whatever).  Are you okay with people distributing in FLAC? 

Why would i not be ok with that?

I meant mp3

[setboy]

You upload a tape online (etree, whatever).  Are you okay with people distributing in FLAC? 

Why would i not be ok with that?

I meant mp3

Yes, that makes more sense now. I would rather people not do other than the archive, but in the end i have the masters and what ever people do with it is what they do with it

[guysonic]

Hard for me to tell as with Flac encoders I've used, seems limited to only 16 bit depth (anyone know of Flac 24 bit, and maximum sample rates over 48K?), but Flac does NOT limit encoded high frequency bandwidth and is supposed to be totally lossless NOT throwing away any audio file information. 

MP3 seems always to more-or-less affect high frequencies encoded bandwidth and also seems limited to 48K sample rates. 

I have used MP3 encoders working directly with 24-to-32bit (float) bit depth files giving 'chance' of more encoded audio details.  But as is generally well known, all MP3 encoders, regardless of encoding (to 320 kbps) rates are NOT lossless, so throw away information to reduce file sizes, and Flac does not do this.

No clear winner here if 24 bit depth MP3 encoding has some benefits over 16 bit Flac limitations.

[Brian Skalinder]

FLAC supports up to 24-bit and up 192 kHz and has for some time, now.

[Teen Wolf Blitzer]

Boojum if you believe all that why do you run a 788, 2 722's, DPA omni's, Schoeps, ect. ( i might add that's a mighty sick setup {minus the shemps}  lol)   Why not run a pair of mics>minidisc?  Do you record in mp3 to the 788?  Just curious.

So because two people couldn't hear the difference that means the rest of the world can't either?

boojum are you a skeptic about everything unless it's in a book somewhere?

Just show me the evidence.  One test is worth a thousand opinions.  I have been down these roads and heard this stuff way longer than you.  That is why I am a skeptic.   ;o)  I have for 55 years heard the stories of Golden Ears and super gear again and again.  It is not impossible, but I am a skeptic.  Listening to your set one time with FLAC/WAV files and the MP3's is not a test.  When someone else plays unknown files for you, on your set if you wish, in an order neither of you know which is which in, that is a test.  And yes, I believe published data way more than the anecdotal BS that floats around unsubstantiated. 

Prove me wrong.  Take a double blind test with a witness and let us see the results.  That would be fair, wouldn't it?    ;o)  You see, it is just that I have heard this stuff for decades but rarely have I seen it pass the rigorous test.  You do not have to believe what I believe but I have the right to believe it and support it.  And if you have a reason that anecdotal evidence is better than rigorous testing, tell us about it.

[notlance]

Boojum if you believe all that why do you run a 788, 2 722's, DPA omni's, Schoeps, ect. ( i might add that's a mighty sick setup {minus the shemps}  lol)   Why not run a pair of mics>minidisc?  Do you record in mp3 to the 788?  Just curious.

A quick answer as to why we should not record directly to a perceptual Codec is because it does not tolerate post processing.  For example, the MP3 Codec may discard low frequencies that are below the threshold of hearing relative to the overall program level.  However, if you wanted to process the initial recording by boosting the bass, there may not be any bass to boost if you recorded in MP3.  Perceptual Codecs may be OK for music distribution, but not for music production.

Chapter 10 of "Principles of Digital Audio" by Ken Pohlmann is a good introduction to perceptual coding.  It is not light reading.

The quality of perceptual codecs depends on many factors: psychoacoustics, implementation, bitrate, listening conditions, listener experience and acuity, etc.  When everything is done right, it can sound pretty good.  The irony is that the main reasons for using perceptual Codecs, namely reducing bandwidth and memory use, are becoming less important.

[rastasean]

Boojum if you believe all that why do you run a 788, 2 722's, DPA omni's, Schoeps, ect. ( i might add that's a mighty sick setup {minus the shemps}  lol)   Why not run a pair of mics>minidisc?  Do you record in mp3 to the 788?  Just curious.

A quick answer as to why we should not record directly to a perceptual Codec is because it does not tolerate post processing.  For example, the MP3 Codec may discard low frequencies that are below the threshold of hearing relative to the overall program level.  However, if you wanted to process the initial recording by boosting the bass, there may not be any bass to boost if you recorded in MP3.  Perceptual Codecs may be OK for music distribution, but not for music production.

Chapter 10 of "Principles of Digital Audio" by Ken Pohlmann is a good introduction to perceptual coding.  It is not light reading.

The quality of perceptual codecs depends on many factors: psychoacoustics, implementation, bitrate, listening conditions, listener experience and acuity, etc.  When everything is done right, it can sound pretty good.  The irony is that the main reasons for using perceptual Codecs, namely reducing bandwidth and memory use, are becoming less important.

So with all this being said about mp3s, what would we be able to do with the lowest bit rate/sampling of a wav file. It must be in some way superior to an mp3.

[rhinowing]

I think you'll find a decent mp3 to be far superior to an 8 bit .wav

[easyed]

FLAC supports up to 24-bit and up 192 kHz and has for some time, now.

other lossless compression formats I like are
APE* http://monkeysaudio.com/
and WavPack** http://wavpack.com/

*Format: PCM WAVE (standard windows .wav file) Sample Rate: anything, Bit depth: 8 or 16 or 24, Channels: 1 or 2
**Compatible with virtually all PCM audio formats including 8, 16, 24, and 32-bit ints; 32-bit floats; mono, stereo, and multichannel; sampling rates from 6 to 192 kHz (and non-standard rates)