Download: FLAC, WAV ?

Shannon Entropy as used in information theory effectively describes the most efficient way to convey information about a source of data.

FLAC encodings will look at repeatable patterns and abbreviate how those pattern are encoded.

.. try FLAC encoding a minute of white noise and a minute of music and a minute of a sine wave  .. the file sizes will be very different... but that is because the source media content is very different and have different predictability ...

A source of data with higher entropy or suprisal will not compress as well in FLAC as opposed to a more predictable one... hence why the very different file sizes between the FLAC encoded noise and sine wave.

By reducing the noise in many source files you are increasing its predictability ... and therefore you are reducing the suprisal and thereby reducing its Shannon Entropy.

Simon

Simon-in-Suffolk posted:

Shannon Entropy as used in information theory effectively describes the most efficient way to convey information about a source of data.

FLAC encodings will look at repeatable patterns and abbreviate how those pattern are encoded.

.. try FLAC encoding a minute of white noise and a minute of music and a minute of a sine wave  .. the file sizes will be very different... but that is because the source media content is very different and have different predictability ...

A source of data with higher entropy or suprisal will not compress as well in FLAC as opposed to a more predictable one... hence why the very different file sizes between the FLAC encoded noise and sine wave.

By reducing the noise in many source files you are increasing its predictability ... and therefore you are reducing the suprisal and thereby reducing its Shannon Entropy.

Simon

 

Simon: Great. Now, in English !

Tech-talk is one thing. It takes an astute/talented chap to relay this into sound quality  interpretations.

Sooo, what are you saying/suggesting ? Are various FLAC encoding/transcoding schemes the reason (even if partially) why some claim FLAC indistinguishable from WAVE, and others not?

pj   

i)  I think the reason why playing flac and wav can sound eupithervthe same or different depending on the renderer used has been explained non-technically several times over in this thread.

ii) As for when the  two formats sound the same to some people but different to others using the same renderer but with other differences in the system - which could include different room acoustics - that is possibly due to the different resolutions of the different systems (or iii below).

iii) and where different people using identical systems, including room, find some can hear differences between file formats while others don't - well, that is their ears/brains.

All assuming double blind testing, or other absence of possible bias. 

pj,

You need to start doing some research into the mathematical nature of information.  However, in summary...

Shanon Entropy is related to how much information is needed to EXACTLY reproduce that specific information set.  Clearly a signal containing only simple sine wave only requires a very limited amount of information to reproduce it exactly, as each sample is precisely related to the one before and the one after it (and can be calculated).  At the other extreme, take a signal composed only of high level white noise ...  You might think, it's only noise so there's no information there - in fact the opposite is true.  Remember the requirement is to reproduce EXACTLY the same information set.  Every different recording of white noise is unique * (as white noise is random) so to reproduce any specific set of white noise requires a lot of information as no sample is related to the one before or the one after it in any way at all.  Every sample has to be individually stored in the sample set.

The difference between FLAC and WAVE in replay isn't to do with the signal encoded in the file (when decoded, the data in a FLAC dile is identical to the data in the equivalent WAVE file).  The difference in replay is due to the engineering nuances in the playback equipment - exactly as IB has pointed out.

* caveat for Simon:  So as to not lose information by simplification in this description; this is qualified as being to the statistical limits of the resolution and encoding of the system used.

HUGE: Hmmm. " You (I) need to start doing some research into the mathematical nature of information."

I see. Thank you for that. Moving quickly,  I'll get to the point. You state: 

"The difference between FLAC and WAVE in replay isn't to do with the signal encoded in the file.." (Yet, this is precisely what some (very talented) data/audio engineers demonstrate is NOT true ! What was once believed (in compression algorithms used for high resolution audio) has been modified/adjusted regularly as more & more critical discoveries are being uncovered. This applies to all areas of hi-f whether analog or digital. Yet, remarkably (actually unsurprisingly) there remain those that establish their understanding based on nothing more than early published data, as if carved in stone -and espouse as much).

You go on to say: "... The difference in replay is due to the engineering nuances in the playback equipment - exactly as IB has pointed out."

And yet, the folks behind Xivero (for example) suggest perhaps that perceived SQ distinctions may ALSO be discovered IN the encoding ? (We are all aware of the variations in the analog chain) 

I can't help but sense that the newish "Streaming" platforms (and software  programs often used in conjunction)  are nothing more than corrupted (oh, I mean exploited) attempts in manipulating/massaging the signal at about the same rate as more and more research unfolds. Could this possibly be the reason why so many passionate audiophiles are constantly seeking ways to "improve" their listening experiences  via a plethora of gear, cables, adapters, programs and poop-loads of other desperate measures ? I bet, there are some real truths in there.           

I seek accurate,  insightful, up-to-date, sophisticated  technical and listening analysis for simple clarity, but most importantly in sharing such "discoveries" so that music/hi-fi lovers can get on with deeply enjoying all that can be relished from their favorite music.

Instead of reciting/re-hashing old news, it's high-time to begin seeking answers to commonly expressed differences. Holding on to firmly entrenched (incomplete science) and espousing its merits is no way moving forward, nor attaining the desired outcome.

pj 

allhifi posted:

HUGE: Hmmm. " You (I) need to start doing some research into the mathematical nature of information."

I see. Thank you for that. Moving quickly,  I'll get to the point. You state: 

"The difference between FLAC and WAVE in replay isn't to do with the signal encoded in the file.." (Yet, this is precisely what some (very talented) data/audio engineers demonstrate is NOT true ! What was once believed (in compression algorithms used for high resolution audio) has been modified/adjusted regularly as more & more critical discoveries are being uncovered. This applies to all areas of hi-f whether analog or digital. Yet, remarkably (actually unsurprisingly) there remain those that establish their understanding based on nothing more than early published data, as if carved in stone -and espouse as much).

You go on to say: "... The difference in replay is due to the engineering nuances in the playback equipment - exactly as IB has pointed out."

And yet, the folks behind Xivero (for example) suggest perhaps that perceived SQ distinctions may ALSO be discovered IN the encoding ? (We are all aware of the variations in the analog chain) 

I can't help but sense that the newish "Streaming" platforms (and software  programs often used in conjunction)  are nothing more than corrupted (oh, I mean exploited) attempts in manipulating/massaging the signal at about the same rate as more and more research unfolds. Could this possibly be the reason why so many passionate audiophiles are constantly seeking ways to "improve" their listening experiences  via a plethora of gear, cables, adapters, programs and poop-loads of other desperate measures ? I bet, there are some real truths in there.           

I seek accurate,  insightful, up-to-date, sophisticated  technical and listening analysis for simple clarity, but most importantly in sharing such "discoveries" so that music/hi-fi lovers can get on with deeply enjoying all that can be relished from their favorite music.

<snip>

You don't currently understand information theory, you don't understand the maths involved, you're not prepared to do the research, and you're not prepared to educate yourself.

Yet when Simon, and I present you with the information about the information theory and the maths, you refuse to accept what we say on the basis of some conspiracy theory "I can't help but sense that the newish "Streaming" platforms (and software  programs often used in conjunction)  are nothing more than corrupted (oh, I mean exploited) attempts in manipulating/massaging the signal at about the same rate as more and more research unfolds.".  Even though Simon, IB, I and other have repeatedly presented you the information that can fully account for all the variances on which those conspiracy theories are based.

allhifi posted:

<snip>

Instead of reciting/re-hashing old news, it's high-time to begin seeking answers to commonly expressed differences. Holding on to firmly entrenched (incomplete science) and espousing its merits is no way moving forward, nor attaining the desired outcome.

pj 

If you want to criticise those of us who do understand, then it's high time you tried to gain sufficient knowledge to understand the information yourself and started to argue from a position of knowledge, rather than relying on conspiracy theory to attack those who do have the understanding and a trying to help you.

Since you quote Xivero as "And yet, the folks behind Xivero (for example) suggest perhaps that perceived SQ distinctions may ALSO be discovered IN the encoding ?"  you need to provide references to back this up; so that we who are capable of understanding the arguments can look and see to what you are referring (and if you've understood it correctly).

HUGE: Please offer up your credentials -as you referenced it now a couple times.

Seriously ? : "... you need to provide references to back this up; so that we who are capable of understanding the arguments can look and see to what you are referring (and if you've understood it correctly).".  WOW. The sheer arrogance.

You fail to recognize, no matter how impressive your level of knowledge may be, only the wisest, honest and indeed smartest realize there's ALWAYS more to learn and understand. With respect to sound-quality observations, one's technical credentials are merely a tool (although an exquisitely important one) that ironically can also reveal its weaknesses -glaringly evident at times.

Moving on, If you scrolled up a few posts (from me specifically) you'd have read a direct explanation by Xivero.com itself; I copied/pasted.  (May I suggest you visit their website).

You continue: " Even though Simon, IB, I and other have repeatedly presented you the information that can fully account for all the variances ..."

" ...to attack those who do have the understanding and a trying to help you."   ????

 Incredible. I await the three Amigo's credentials before we can move on.

pj

nbpf posted:

allhifi posted:

...

The astonishing thing is that this can be clearly revealed through an outboard, decent (but cheap) 3-way desktop speaker system ! (How in the world (when you think about it) could this analog representation steeped in basic loudspeaker performance, poor speaker placement (upon my computer desk; Left speaker 12-16" behind my monitor, although angled toward me lol, and the Right channel speaker nestled between -and slightly behind- the monitor and a massive Canon printer extending way past the front plane of the speaker face)  offer up such SQ distinctions within all of the analog stages using a precise digital file rendering (of all the formats) ?

Amazing.

...

It is the Canon printer! It has been hacked by hard boiled pro-wav-files evengelists. They have modified the printer's firmware and now your device sends high-frequency electromagnetic beams every time it detects a .flac or a .aiff file on its way to your dac. The beams are tuned to deteriorate sound quality. I suggest random renaming of .flac and .aiff files to .wav and vice versa. Or switching off the printer. Best, nbpf 

My HiCap DR actually starts to make an audible hum when I print on my Samsung Printer.

Hmmmm ... what do you recommend, then ? (lol)

pj

Actually, it appears you have a ground-loop hum induced issue.

pj 

Ardbeg10y posted:

nbpf posted:

allhifi posted:

...

The astonishing thing is that this can be clearly revealed through an outboard, decent (but cheap) 3-way desktop speaker system ! (How in the world (when you think about it) could this analog representation steeped in basic loudspeaker performance, poor speaker placement (upon my computer desk; Left speaker 12-16" behind my monitor, although angled toward me lol, and the Right channel speaker nestled between -and slightly behind- the monitor and a massive Canon printer extending way past the front plane of the speaker face)  offer up such SQ distinctions within all of the analog stages using a precise digital file rendering (of all the formats) ?

Amazing.

...

It is the Canon printer! It has been hacked by hard boiled pro-wav-files evengelists. They have modified the printer's firmware and now your device sends high-frequency electromagnetic beams every time it detects a .flac or a .aiff file on its way to your dac. The beams are tuned to deteriorate sound quality. I suggest random renaming of .flac and .aiff files to .wav and vice versa. Or switching off the printer. Best, nbpf 

My HiCap DR actually starts to make an audible hum when I print on my Samsung Printer.

You see? Everybody seems to be scared by self-driving cars, mobile phones and biometric data leaks. Let me tell you: it's these other humble servants that are going to lead the next revolution and break our necks!  Printers and refrigerators, core transformers, heating pipes, elevators. They have served us silently for decades or even for centuries. They are now fed up with our lack of respect and care. They are fed up with our flippant flirting with smart devices! They will punish Allhifi, they will punish all of us, I'm afraid.    

It made me consider that a mathematical function which describes the the 50hz curve (like flac) is much more efficient than recording it in raw data (like wav).

allhifi posted:

HUGE: Please offer up your credentials -as you referenced it now a couple times.

Seriously ? : "... you need to provide references to back this up; so that we who are capable of understanding the arguments can look and see to what you are referring (and if you've understood it correctly).".  WOW. The sheer arrogance.

You fail to recognize, no matter how impressive your level of knowledge may be, only the wisest, honest and indeed smartest realize there's ALWAYS more to learn and understand. With respect to sound-quality observations, one's technical credentials are merely a tool (although an exquisitely important one) that ironically can also reveal its weaknesses -glaringly evident at times.

Moving on, If you scrolled up a few posts (from me specifically) you'd have read a direct explanation by Xivero.com itself; I copied/pasted.  (May I suggest you visit their website).

You continue: " Even though Simon, IB, I and other have repeatedly presented you the information that can fully account for all the variances ..."

" ...to attack those who do have the understanding and a trying to help you."   ????

 Incredible. I await the three Amigo's credentials before we can move on.

pj

 

Of specific relevance, amongst other areas, I have worked professionally at a design level in
Electronic Engineering
Computer interface Hardware systems design
Computer Systems design (interfacing)
Computer Systems design (mathematical modelling including algorithm development)
Computer Systems design (Data Transmission systems)

In a non-professional capacity I've designed and built HiFi amps.

And what is your relevant design experience?

So are you going to move on now?

The piece you quoted from Xivero.com is totally irrelevant as that applies to lossy compression "A FLAC encoder becomes much more efficient if we reduce the entropy (noise) of an audio file before it gets compressed."  The action of removing the noise is a lossy transformation.

allhifi posted:

.

Moving on, If you scrolled up a few posts (from me specifically) you'd have read a direct explanation by Xivero.com itself; I copied/pasted.  (May I suggest you visit their website).

 

 

Sadly, it seems that you either can't, or won't read and digest what is said in many of the posts that have been offered to help you, the OP, understand the questions you raise. I partly base my statement on the fact that after I stated things several times over, trying to put it in a way you understood, you requested me to do so agaian, even more simply - then some hours later said you had read it again and did understand (though I remain unconvinced).

I have no reason to believe or disbelieve anything said on the xivero.com website you cite, however, My question to you is, when you read something on a website, what efforts do you take to verify that what you read is fact? Bear in mind that any commercial website exists primarily to help the business sell its wares and it is far from unknown for 'facts' to be less than entirely factual, and often are at the very least influenced by marketing hype, and sometimes more hype than fact. Be clear, I am NOT saying that xivero.com presents anything other pure objective and genuine facts - but I am also not saying that it is not all pure hype: the question is, when you read things like they say, how do YOU decide which it is?

Excellent. Thank for that. And so, (after listing your qualifications & experience) that assures you that : " Even though Simon, IB, I and other have repeatedly presented you the information that can fully account for all the variances ..."

( If you are right -that you three (and others) have offered up information "THAT CAN FULLY ACCOUNT FOR ALL THE VARIANCES"  -I owe the 'collective' an apology.  If you're wrong, firther discussions (and revelations) must follow.  

HUGE: FLAC is a lossless file-format. What are you saying ??   Good God, this is getting better by the minute.

pj

allhifi posted:

Excellent. Thank for that. And so, (after listing your qualifications & experience) that assures you that : " Even though Simon, IB, I and other have repeatedly presented you the information that can fully account for all the variances ..."

( If you are right -that you three (and others) have offered up information "THAT CAN FULLY ACCOUNT FOR ALL THE VARIANCES"  -I owe the 'collective' an apology.  If you're wrong, firther discussions (and revelations) must follow.  

" And what is your relevant design experience ?"  NONE. 

" So are you going to move on now ?"  Assuming accuracy, yes. 

The piece you quoted from Xivero.com is totally irrelevant as that applies to lossy compression (FLAC is lossy ?) "A FLAC encoder becomes much more efficient if we reduce the entropy (noise) of an audio file before it gets compressed."  The action of removing the noise is a lossy transformation. Are you kidding me ?

HUGE: FLAC is a lossless file-format. What are you saying ??   Good God, this is getting better by the minute.

pj

Like Reply (0 Likes)

 

Flac is lossless.

Remove noise AND encode as FLAC is NOT lossless.

The action of removing the noise is a lossy transformation. Are you kidding me ?
          - No I'm not kidding you: Removing noise IS a lossy transformation.

nbpf posted:

Ardbeg10y posted:

nbpf posted:

allhifi posted:

...

The astonishing thing is that this can be clearly revealed through an outboard, decent (but cheap) 3-way desktop speaker system ! (How in the world (when you think about it) could this analog representation steeped in basic loudspeaker performance, poor speaker placement (upon my computer desk; Left speaker 12-16" behind my monitor, although angled toward me lol, and the Right channel speaker nestled between -and slightly behind- the monitor and a massive Canon printer extending way past the front plane of the speaker face)  offer up such SQ distinctions within all of the analog stages using a precise digital file rendering (of all the formats) ?

Amazing.

...

It is the Canon printer! It has been hacked by hard boiled pro-wav-files evengelists. They have modified the printer's firmware and now your device sends high-frequency electromagnetic beams every time it detects a .flac or a .aiff file on its way to your dac. The beams are tuned to deteriorate sound quality. I suggest random renaming of .flac and .aiff files to .wav and vice versa. Or switching off the printer. Best, nbpf 

My HiCap DR actually starts to make an audible hum when I print on my Samsung Printer.

You see? Everybody seems to be scared by self-driving cars, mobile phones and biometric data leaks. Let me tell you: it's these other humble servants that are going to lead the next revolution and break our necks!  Printers and refrigerators, core transformers, heating pipes, elevators. They have served us silently for decades or even for centuries. They are now fed up with our lack of respect and care. They are fed up with our flippant flirting with smart devices! They will punish Allhifi, they will punish all of us, I'm afraid.    

Oh my. lol

pj

(P.S. I'm already taking security precautions. Thank you. )

Huge posted:

allhifi posted:

Excellent. Thank for that. And so, (after listing your qualifications & experience) that assures you that : " Even though Simon, IB, I and other have repeatedly presented you the information that can fully account for all the variances ..."

( If you are right -that you three (and others) have offered up information "THAT CAN FULLY ACCOUNT FOR ALL THE VARIANCES"  -I owe the 'collective' an apology.  If you're wrong, firther discussions (and revelations) must follow.  

" And what is your relevant design experience ?"  NONE. 

" So are you going to move on now ?"  Assuming accuracy, yes. 

The piece you quoted from Xivero.com is totally irrelevant as that applies to lossy compression (FLAC is lossy ?) "A FLAC encoder becomes much more efficient if we reduce the entropy (noise) of an audio file before it gets compressed."  The action of removing the noise is a lossy transformation. Are you kidding me ?

HUGE: FLAC is a lossless file-format. What are you saying ??   Good God, this is getting better by the minute.

pj

Like Reply (0 Likes)

 

Flac is lossless.

Remove noise AND encode as FLAC is NOT lossless.

The action of removing the noise is a lossy transformation. Are you kidding me ?
          - No I'm not kidding you: Removing noise IS a lossy transformation.

I'm no Einstein, nor am I HUGE, but unless there is some data buried within the noise (previously unexpected), this seems reasonable. Otherwise, how in the world can noise removal (of a music data stream) be considered "lossy"?

Enlighten me -even if that means you must educate me via an appropriate/relevant link.

pj 

( I will offer up your thoughts to others in the field that share your (claimed) credentials to chime in. I'm simply using logic in my arguments/questions.)     

allhifi posted:

Huge posted:

Flac is lossless.

Remove noise AND encode as FLAC is NOT lossless.

The action of removing the noise is a lossy transformation. Are you kidding me ?
          - No I'm not kidding you: Removing noise IS a lossy transformation.

I'm no Einstein, nor am I HUGE, but unless there is some data buried within the noise (previously unexpected), this seems reasonable. Otherwise, how in the world can noise removal (of a music data stream) be considered "lossy"?

Enlighten me -even if that means you must educate me via an appropriate/relevant link.

pj 

( I will offer up your thoughts to others in the field that share your (claimed) credentials to chime in. I'm simply using logic in my arguments/questions.)     

If you look at the explanation of Shannon Entropy you'll see that noise is still information and its removal is thus lossy.

Here it is again...

Shanon Entropy is related to how much information is needed to EXACTLY reproduce that specific information set.  Clearly a signal containing only simple sine wave only requires a very limited amount of information to reproduce it exactly, as each sample is precisely related to the one before and the one after it (and can be calculated).  At the other extreme, take a signal composed only of high level white noise ...  You might think, it's only noise so there's no information there - in fact the opposite is true.  Remember the requirement is to reproduce EXACTLY the same information set.  Every different recording of white noise is unique * (as white noise is random) so to reproduce any specific set of white noise requires a lot of information as no sample is related to the one before or the one after it in any way at all.  Every sample has to be individually stored in the sample set.

To look at it another way, deliberately very simplistically: in a recorded audio signal, how do you know what is noise? Is it that feint backgoind hiss you might hear? But what is that hiss, and how exactly do you distinguish when a cymbal decay has ceased? Remove the hiss, and you may remove part of the cymbal sound that was dying away in there. Is the noise actually the ambiance of the recording venue? Remove it and you've removed or changed the ambiance. 

It may be valid for the mastering engineer to remove what he or she determines is sound not wanted in the released recording (whether or not noise), but removing it subsequently by some algorith is modifying the recording. Nothing wrong in that if it is deliberate and the person so doing decides they like it, but what results isn't the same as the recording.

I have one recording that apparently has a background hiss.  Except it isn't: It's the sound of a rain shower falling outside the hall during the recording.

Do you remove that part of the signal?

I.B.: Nicely said.

Realizing you have offered up as an example only : " and how exactly do you distinguish when a cymbal decay has ceased? ", surely this has been determined (with 99% accuracy) via endless research into the topic. It's rather simple to recognize the 'crash cymbal's spectral profile extending beyond 40 KHz. (and beyond) where it eventually becomes submerged in noise. Other recordings have demonstrated content in the 50-60 KHz. range. How can it be difficult to ascertain signal from noise ?

You bring up an interesting  point: " Is the noise actually the ambiance of the recording venue? Remove it and you've removed or changed the ambiance." 

 Recording space or hall venue acoustics remain (I suspect) in the the very low frequency range (0-20/30 ? Hz.)  My guess is that the 'noise' often referred to is much higher in frequency. If not known already, surely soon we shall have the answers you present. And, in that regard, HUGE's current understanding/interpretation may very well become modified as a result -as  all others, naturally.    

pj

Huge posted:

allhifi posted:

Huge posted:

Flac is lossless.

Remove noise AND encode as FLAC is NOT lossless.

The action of removing the noise is a lossy transformation. Are you kidding me ?
          - No I'm not kidding you: Removing noise IS a lossy transformation.

I'm no Einstein, nor am I HUGE, but unless there is some data buried within the noise (previously unexpected), this seems reasonable. Otherwise, how in the world can noise removal (of a music data stream) be considered "lossy"?

Enlighten me -even if that means you must educate me via an appropriate/relevant link.

pj 

( I will offer up your thoughts to others in the field that share your (claimed) credentials to chime in. I'm simply using logic in my arguments/questions.)     

If you look at the explanation of Shannon Entropy you'll see that noise is still information and its removal is thus lossy.

Here it is again...

Shanon Entropy is related to how much information is needed to EXACTLY reproduce that specific information set.  Clearly a signal containing only simple sine wave only requires a very limited amount of information to reproduce it exactly, as each sample is precisely related to the one before and the one after it (and can be calculated).  At the other extreme, take a signal composed only of high level white noise ...  You might think, it's only noise so there's no information there - in fact the opposite is true.  Remember the requirement is to reproduce EXACTLY the same information set.  Every different recording of white noise is unique * (as white noise is random) so to reproduce any specific set of white noise requires a lot of information as no sample is related to the one before or the one after it in any way at all.  Every sample has to be individually stored in the sample set.

How does that differ  from what Xivero is saying ?

The methods FLAC applies for lossless compression are based on the technology of “Sparse Sampling of Signals with Finite Rate of Innovation”.

A piece of an audio signal is approximated either by a simple polynomial or a linear predictive coding (LPC).

An audio sample sequence containing noise is not really a special signal of finite rate of innovation (FIR) that could be sparsely sampled without taking care about the residual error and therefore it is not enough to just encode the coefficients of the polynomial but also the residual error!

The residual error increases with the entropy of the data to be compressed.

pj

Huge posted:

I have one recording that apparently has a background hiss.  Except it isn't: It's the sound of a rain shower falling outside the hall during the recording.

Do you remove that part of the signal?

Absolutely NOT ! The high-humidity atmosphere may have very well damped some listening room/recording venue space "resonances"  -and therefore, must be retained ! Yes, I'm kidding. Good point. But. you were unable to determine rain (natural acoustic) from unwanted system/recording noise ?

pj 

It's no different, but they only claimed that the FLAC part of the process is lossless.

The whole process is still lossy.  Using that whole process to try to prove anything about FLAC is utterly irrelevant.

The losses involved in removing the "noise" invalidate any conclusion form comparison of the result of the subsequent FLAC encoded data to the original LPCM data.  To make any such comparison is fundamentally flawed even in principle never mind in execution.

The data compression in FLAC is functionally equivalent to the data compression of a ZIP file, but optimised for music.  If you compress the data into a FLAC file then decompress the data, the data that result are EXACTLY the same as you put into the FLAC file in the first place.  The presence or absence of noise makes no difference to that.  However if you add a pre-step of removing the noise you're also removing part of the signal which you can NEVER recover - and that makes your whole process lossy.

If you remove the noise you invalidate the comparison..
If you leave the noise untouched, then there's no difference between the LPCM data, the data decoded from the WAVE file and the data decoded from the FLAC file.

It really is that simple.