Is It Time To Rethink Lossless?

Perceptionally lossless?😎

I like your train of thought here.  It is imperative to have common ground, even more so as new formats come and go.  We do not need another ongoing corporate controversy.

I know that delivery systems are in flux right now.   How does that change things?  What if Apple figures out how to deliver TrueHD?  (I can dream)

Can we get this down to one descriptive sentence or two?  The old elevator speech thing...  

From my point of view, the term "lossless" or "bitperfect" is relative to the "transport from the source" before arriving at the decoding point. And yes, the logic inherits that sources could be perceptionally lossless or lossy, too.
That's why the source and often also the provenance are important for many of us.
Thus, it's nothing worth without a quality transport (lossless, bitperfect) and decoding. The transport (stream / streamer / audio transport) in my opinion should be as transparent, free of alteration and noise, bitperfect and lossless as it could be.
If you call it "lossless" or "ducktolled" doesn' matter for me, as long as it is as close to the source quality I bought once or will buy in the future.
 

I think my conception of lossy vs lossless is a bit different. Yes, mastering, mixing, sample rate conversion would all cause changes in the integrity of the original recorded signal and have some sonic effects like subtle losses of timing/transient accuracy or subtle losses of detail.

But to me, a true lossy compression scheme, be it mp3 or Dolby Digital Plus or MQA generally have ways to purposely "throw out" sonic information based on psychoacoustic models of hearing thresholds and /or sound masking. Even though they are meant to be perceptually lossless, for me, as an audiophile who loves listening to orchestral classical music, I often find there is always an obvious loss of subtle details if you're listening for it, and much more significant changes to timing/transient accuracy.

To me, the bigger question at hand it sounds like is that if you have a master that was in lossless Dolby Atmos, is it better to hear it in lossy Dolby Digital Plus Atmos with the preserved spatial audio design or is it better to hear it in lossless stereo where you lose the spatial audio information that was originally intended in the master. I presume that'll depend on the music and the listener. Because the philosophical question becomes: are stereo recordings "lossy" because it doesn't capture the spatial audio information.

That said, I listen to music for music first. So I'd always prefer lossy music to no music at all.

6 minutes ago, ecwl said:

I often find there is always an obvious loss of subtle details if you're listening for it, and much more significant changes to timing/transient accuracy.

This is of course if you have a known version that's then converted using a lossy codec. I'm right there with you. For example, a CD to 16/44.1 WAV to 128 kbps MP3. 

When there is no original source or we don't know which version is considered the original, things get sticky. 

11 minutes ago, ecwl said:

To me, the bigger question at hand it sounds like is that if you have a master that was in lossless Dolby Atmos, is it better to hear it in lossy Dolby Digital Plus Atmos with the preserved spatial audio design or is it better to hear it in lossless stereo where you lose the spatial audio information that was originally intended in the master. I presume that'll depend on the music and the listener. Because the philosophical question becomes: are stereo recordings "lossy" because it doesn't capture the spatial audio information.

I certainly hear you on this one, and thank you for bringing it up. I've done this listening countless times in the last couple years. As a music loving audiophile, I want the "best" version! 

You are 100% correct in that it depends on the music and the listener, and I'll add the playback system. I've even hear albums that sound best, to me, on headphones in Atmos. Yes, the headphone Atmos mix was better than the speaker Atmos and all of the high resolution stereo mixes. Crazy. 

The theoretical question about stereo being lossy without the spatial info is interesting as well. What I've found is that stereo sounds synthetically crammed into the confines to two channels compared to an immersive soundstage. So, in a way it isn't lossy because all the elements are present, but they are all coming from the front two channels. 

I look at it this way: stereo and Atmos are different ways to experience the wonders of music. As much as I love Atmos, I would never advocate for it to replace stereo because I know people that love stereo more than anything. 

I could discuss this stuff for days. It's interesting and I value all the perspectives. 

11 minutes ago, ecwl said:

That said, I listen to music for music first. So I'd always prefer lossy music to no music at all.

Absolutely. In addition, I'll take my favorite music on an AM radio over something I dislike at 24/384. 

48 minutes ago, bbosler said:

 

this is a good one. It is unknown but it is highly doubtful....  All of the sudden one day Amazon announces streams available at  24/192  BETTER THAN CD 

 

when it came out I saw Led Zeppelin titles as 24/192 and though GREAT !! But then I thought, where the hell did they get original 24/192 files for stuff that was never recorded that way? So I bought one and downloaded it , or at least tried to,  only to find out that they won't allow you to  download the 24/192 files. At least they wouldn't. I never tried after the initial try

 

In any case, many if not most are  obviously upsampled. Like HDtracks got caught doing early on and maybe still does. I gave up on them too.

 

So is upsampled  lossless? Can the original 16/44.1 be reconstructed from a 24/192 upsample? I doubt it. And now that I think about it, where did the 16/44.1 files come from? Some originally converted  from 16/48  DAT tapes

 

In the case of Led Zeppelin, I think the "originals" would be analog tapes which can be digitized at any resolution supported by the A/D converter. 

1 hour ago, kumakuma said:

In the case of Led Zeppelin, I think the "originals" would be analog tapes which can be digitized at any resolution supported by the A/D converter. 

agreed, they can be, but were they?

HDtracks has them at 24/96 from 2014, but they never say where they came from either. If original 24/192 existed it seems they would be  available from other than Amazon streaming,  so what are the chances that Amazon  got 24/192 files of the  tapes? I'm going with zero.

I used to fret on these distinctions myself for the longest time. Now I just call it HiFi because of the fanciness in the playback chain, not the source media itself. I would go crazy trying to guess how the original one was made.

I sort of also learned not to trust online music stores, even the audiophile ones. Case in point, I remember for us rock fans it was kind of a big thing when Metallica's black album was released in "HiRes" from a store I will not name. In forums most agreed the CD sounded better; some dug deeper and apparently the HiRes 24/96 FLAC was based on the already diluted one made for the 5.1 from the A-DVD that was sold some time ago.

So what's the true source? It all comes down to this, right?

The only thing I can think of is that Studio's make the master recording information public, if that's the one the artist/ Studio is releasing, and have it displayed in the streaming services and such.

Or maybe showing a screenshot of the Pro Tools working file? 😉

3 hours ago, bbosler said:

Can the original 16/44.1 be reconstructed from a 24/192 upsample? I doubt it.

It depends on a few things so it's not something impossible in principle:

# upsample
sox "02. Fast Car.flac" -b24 "up.wav" rate -s 192000

# downsample
sox -D "up.wav" -b16 "down.wav" rate -s 44100

# compare (subtract one from the other)
sox -m -v 1 "02. Fast Car.flac" -v -1 "down.wav" -n stats
             Overall     Left      Right
..
Pk lev dB       -inf      -inf      -inf
RMS lev dB      -inf      -inf      -inf

Anything that has an analogue form can be considered lossy - even the very best master, analogue tape, as per the direct capture of the microphones, is lossy - because the tape material degrades over time. Only anything digital can be considered lossless - because a perfect replica can be recreated at any time, ad infinitum, over as long a time period as you want; even the worst MP3 version of something is lossless in this sense.

If one deliberately discards digital information from some, 'master', digital version, in a conversion, then this is also lossy, in an absolute sense. The question then is whether anyone can hear a significant difference between the original, and the 'lossy' variant, under any circumstances. The final touch would be, can the lossy digital variant be converted back to the format of the original, using best methods, and the original and the reconstructed lossy copy then be indistinguishable under any listening conditions - if so, then IMO the 'lesser' version is indeed lossless.

1 hour ago, danadam said:

It depends on a few things so it's not something impossible in principle:

# upsample
sox "02. Fast Car.flac" -b24 "up.wav" rate -s 192000

# downsample
sox -D "up.wav" -b16 "down.wav" rate -s 44100

# compare (subtract one from the other)
sox -m -v 1 "02. Fast Car.flac" -v -1 "down.wav" -n stats
             Overall     Left      Right
..
Pk lev dB       -inf      -inf      -inf
RMS lev dB      -inf      -inf      -inf

 

same peak and rms levels does not mean the files are identical.

I don't see how it is mathematically possible to go from 16/44.1 to 24/192 and back to 16/44.1 and have a file  identical to the original.

to get a 24 bit sample 192,000 times a second in between 16 bit samples 44,100 times a second , you have to extrapolate (estimate) what those levels are. Then you have to extrapolate (estimate) 16 bit levels 44,100 times each second. Especially since the rates are not evenly divisible, you just can't get that to be exactly the same in each direction 

21 minutes ago, bbosler said:

same peak and rms levels does not mean the files are identical.

Sure, but that's not what the example is showing. It shows that the result of subtracting the up- and downsampled version from the original file is a digital silence. This means that each and every sample in the up- and downsampled version is the same as in the original file.

4 minutes ago, danadam said:

Sure, but that's not what the example is showing. It shows that the result of subtracting the up- and downsampled version from the original file is a digital silence. This means that each and every sample in the up- and downsampled version is the same as in the original file.

In essence, it’s like you added a paragraph to the end of a text file then removed it. Not that your example is wrong, it’s just a very simple, if not unrealistic, exercise. 

21 minutes ago, The Computer Audiophile said:

Not that your example is wrong, it’s just a very simple, if not unrealistic, exercise. 

I believe the example is definitely wrong. Peak and rms is not the equivalent of comparing each and every sample. 

if you were using integer multiples (like 48K to 192K) and just padding 16 to 24 bits with zeros, then yes. But 44.1K to 192K is different since the 192K samples are in between the 44.1K samples at odd intervals. The program has to extrapolate and round off going in both directions. How can it possibly end up exactly where  it started?

doesn't most playback software recommend even integer conversions for this very reason? That going from 44.1 to 192 instead of 44.1 x 4 to  176.4 introduces errors?

i am willing to admit I am wrong, but peak and rms levels is no proof. 

These days we can always use AI to fill in the missing information!

I'm sticking with the traditional definition.  To me it means a bit-perfect recreation of the original source whatever that source was.

6 hours ago, Audiophile Neuroscience said:

WAV is lossless because it was never compressed

Lossless to what? What about an album that available as 24/192 and 16/44.1 WAV, where high frequency content from the 24/192 version couldn’t be contained in the CD quality version? Both WAV files. 
 

6 hours ago, Audiophile Neuroscience said:

So, is it too simplistic to think that lossless means no loss of information due to a compression process and relative to the source being compressed?

What source? 
 

Is loss of information due to resampling OK?

Are CDs lossless, even though by definition they can’t be if the source contained information above 22.05 kHz?

I’m not asking to be confrontational, just asking to see where you loosen the definition from its strict meaning. All with good intentions and interest in your answers. 

13 hours ago, The Computer Audiophile said:

In essence, it’s like you added a paragraph to the end of a text file then removed it. Not that your example is wrong, it’s just a very simple, if not unrealistic, exercise.

I thought that was the premise of the question, can you get back the 16/44 original from the 24/192 upsample. The assertion was (and apparently still is) that it's never possible, I showed that it sometimes is. I can agree that steep anti-imaging filter (99% bandwidth, the "-s" option) is probably not something that would be usually used, but I'm not sure if that's what you meant by calling it simple.

13 hours ago, bbosler said:

I believe the example is definitely wrong. Peak and rms is not the equivalent of comparing each and every sample. 

Could you explain what makes you think that I only compared peak and rms values of the 2 files? Because once again, that operation did compare each and every sample. Maybe if I split it into separate commands it will be more visible:

# create null file
sox -m -v 1 "02. Fast Car.flac" -v -1 "down.wav" "null.wav"

# show statistics of the null
sox "null.wav" -n stats
             Overall     Left      Right
..
Pk lev dB       -inf      -inf      -inf
RMS lev dB      -inf      -inf      -inf

I can also decompress the flac to wav and compare the files directly:

flac -d "02. Fast Car.flac"

md5sum "02. Fast Car.wav" "down.wav"
efcb88d3e5be874902f03bddac953086  02. Fast Car.wav
efcb88d3e5be874902f03bddac953086  down.wav

diff -s "02. Fast Car.wav" "down.wav"
Files 02. Fast Car.wav and down.wav are identical