If you had to choose...

[Jud]

Admittedly, dithering might be a culprit, but on the other hand, upsampling does not seem to loose any data. If anything, it adds additional data.

 

I just converted a small 16/44.1 file to 16/48.0, and was able to find and identify identical samples in the resampled file for each sample in the in the original file file. (AIFF).

 

I would agree the conversion might not be perfect, but I do not think it is "lossey."

 

Yours,

-Paul

 

Leaving aside completely anything about "a ton of ultrasonic noise," which is part of a controversy about DSD I very much do not think should become part of this thread, yes, Paul, it is lossy - **you cannot reconstruct the original file from the resampled one**. You won't get a perfect data match for the original with any mathematical or filtering operation you try. So yes, in terms of a strict definition, it is indeed lossy, lossless being defined as something from which you can, with mathematical perfection, reconstruct the original.

[ringenesherre]

Paul,

 

basically PCM -> DSD has to be lossy, as DSD -> PCM is lossy.

 

I'm sure we can agree on the truth of the statements A and B below.

 

A) Lossy is an antonym of lossless.

 

B) Lossless means that when converting from X to Y, all information from X can be recovered knowing only Y.

 

Obviously, PCM -> DSD is lossless if, and only if, DSD -> PCM is lossless.

[Teresa]

If you had to choose only one format to rip all your cd library, which would it be?

 

DSD, pref .dsf for the metadata.

 

1+ My choice also.

 

Oops! I misread the question I thought the original poster was wanting us to choose only one format for our library. I must have read over the "CD" part. Since I don't own any CDs or for that matter any 16/44.1 music files I must withdraw my answer.

 

I'm a bit concerned about all the "DSD / DSF" responses.

 

Sorry, I responded incorrectly since I don't own any CDs and my lowest resolution music file is 24/88.2kHz PCM. I selected DSD as my one choice for my library as I'm not willing to convert my best sounding music files (DSD 2.8MHz and 5.6MHz) to any PCM music format.

 

Currently I have all my music files in their native format (PCM and DSD).

 

• I have DSD music files in dsf and dff
  
• I have AIFF for Pure Music/iTunes on my Mac Mini
  
• I have WAV for the Teac HR Audio player on my Mac Mini and for Foobar on my Gateway laptop.
  
• 
  

 

The WAV and AIFF files are duplicates of each other and I have been thinking of converting them to uncompressed FLAC, what do you think?

 

Or should I convert my high resolution PCM files to DSD and should I choose 2.8 or 5.6?

[Paul R]

Leaving aside completely anything about "a ton of ultrasonic noise," which is part of a controversy about DSD I very much do not think should become part of this thread, yes, Paul, it is lossy - **you cannot reconstruct the original file from the resampled one**. You won't get a perfect data match for the original with any mathematical or filtering operation you try. So yes, in terms of a strict definition, it is indeed lossy, lossless being defined as something from which you can, with mathematical perfection, reconstruct the original.

 

Um- I am not so sure of this Jud. I will have to think on it. I do not think being able to reconstruct the original file serves as a definition for lossless, but of course, it is one common test to determine if a process is lossless.

 

I believe, mathematically, you certainly could reconstruct the original file from the DSD data, but you would be decimating the data to do that. That is, I think we all agree, a "lossy" operation. Even if you wound up with exactly the same original file. This is, the same thing as upsampling PCM data from 16/44.1 to 24/88.2. No data is lost in the upsample process, but it is tricky to get the original file back from the up sampled data.

 

Going from PCM -> DSD you are not in any way discarding any of the data, and thus it is not lossy.

 

Skeptic - adding data to a file via interpolation is not a "lossy" operation. At least any more lossy than PCM, which pretty much does exactly that at playback time.

 

Like I said, I may just be dense on this subject, but lossy to me means data has been discarded, as when decimating a data file in the process of converting a DSD128 file to PCM 24/192. The target file simply cannot represent all the data that is in the original file.

 

-Paul

[jhwalker]

Um- I am not so sure of this Jud. I will have to think on it. I do not think being able to reconstruct the original file serves as a definition for lossless, but of course, it is one common test to determine if a process is lossless.

 

I believe, mathematically, you certainly could reconstruct the original file from the DSD data, but you would be decimating the data to do that. That is, I think we all agree, a "lossy" operation. Even if you wound up with exactly the same original file. This is, the same thing as upsampling PCM data from 16/44.1 to 24/88.2. No data is lost in the upsample process, but it is tricky to get the original file back from the up sampled data.

 

Going from PCM -> DSD you are not in any way discarding any of the data, and thus it is not lossy.

 

Skeptic - adding data to a file via interpolation is not a "lossy" operation. At least any more lossy than PCM, which pretty much does exactly that at playback time.

 

Like I said, I may just be dense on this subject, but lossy to me means data has been discarded, as when decimating a data file in the process of converting a DSD128 file to PCM 24/192. The target file simply cannot represent all the data that is in the original file.

 

-Paul

 

IMO, if you cannot exactly reconstruct the original content (e.g., ZIP > unzipped files), the process is "lossy".

 

After all, you wouldn't say a theoretical ZIP file that returned *close to* the same values in your ZIP'd spreadsheet was "lossless", would you?

 

So any process that cannot exactly reconstruct the original content (e.g., FLAC > WAV < ALAC > AIFF, et al.) cannot be considered lossless. That's the way I think about it, at least.

 

So my methodology would be to rip / archive my data in its native format (for SACD, that would be DSD, for CD, PCM) at the same rate / bit depth originally recorded. If, after that, you want to upsample, convert, etc., you can do so to your heart's content, but you still have that original bit perfect copy.

[Jud]

Um- I am not so sure of this Jud. I will have to think on it. I do not think being able to reconstruct the original file serves as a definition for lossless, but of course, it is one common test to determine if a process is lossless.

 

I believe, mathematically, you certainly could reconstruct the original file from the DSD data, but you would be decimating the data to do that. That is, I think we all agree, a "lossy" operation. Even if you wound up with exactly the same original file. This is, the same thing as upsampling PCM data from 16/44.1 to 24/88.2. No data is lost in the upsample process, but it is tricky to get the original file back from the up sampled data.

 

Going from PCM -> DSD you are not in any way discarding any of the data, and thus it is not lossy.

 

 

-Paul

 

Paul, any interpolation filter used in audio may happen to output some of the identical values as the inputs, but that fact doesn't allow mathematical reconstruction. You're certainly welcome to check this out at any source you like, and you will find that you will never be able to reconstruct the original with mathematical precision from any interpolation-filtered file. It appears to me from using HQPlayer that indeed converting PCM to DSD does involve interpolation filtering, so I think that's our answer.

[Jud]

The WAV and AIFF files are duplicates of each other and I have been thinking of converting them to uncompressed FLAC, what do you think?

 

Or should I convert my high resolution PCM files to DSD and should I choose 2.8 or 5.6?

 

Teresa, I'm not expressing any opinion re uncompressed FLAC vs. WAV and AIFF. But regarding conversion to DSD, I'd recommend against it if the purpose is for *archiving*. The reason is that if there is better DSD conversion developed in the future using the more powerful computers we'll presumably have then, you won't be able to take advantage of it, having lost the original WAV and AIFF files. If it is purely for playback right now, then just do what your ears tell you of course.

[miguelito]

Aiff

[miguelito]

So my methodology would be to rip / archive my data in its native format (for SACD, that would be DSD, for CD, PCM) at the same rate / bit depth originally recorded. If, after that, you want to upsample, convert, etc., you can do so to your heart's content, but you still have that original bit perfect copy.

Completely agree.

[Paul R]

Paul, any interpolation filter used in audio may happen to output some of the identical values as the inputs, but that fact doesn't allow mathematical reconstruction. You're certainly welcome to check this out at any source you like, and you will find that you will never be able to reconstruct the original with mathematical precision from any interpolation-filtered file. It appears to me from using HQPlayer that indeed converting PCM to DSD does involve interpolation filtering, so I think that's our answer.

 

We must be talking about some different kind of interpolation.

 

Typically, one would up-sample the signal, and basically that is just stuffing zero value samples between the original samples. Adding more samples per time period is equivalent to increasing the sampling rate.

 

Interpolation - at least in the sense used with DSP, not pure mathematics - is up-samplng followed by filtering, basically to get rid of the high frequency effects. Using something like a FIR filter is computationally efficient, especially if broken into stages. But the end result is the same and you wind up with the original samples and nicely computed extra samples between them. The original data is preserved.

 

So yes, at least in theory, you could retrieve the original data, or at least, I don't see why you could not, given the original upsampling algorithm. All that is necessary is to identify the original samples from the generated ones. (I say that knowing it is not quite that simple. But it is fact, what decimation is meant to do.)

 

I may be wrong here, and if so, it will be terribly embarrassing, but I don't think so.

 

Now, when you transcode to DSD, that's a bit of a different animal, since you are not only changing the format of the data, but the meaning too. And dealing with stuff like byte order processing. Still, it is essentially the same kind of process. You are just converting a sample value into a bit stream. Going back may be a little more difficult, but I still see no reason why it would be impossible.

 

And of course, I disagree that you have to be able to recreate the original file for a reproduction to be lossless. If no information is lost, then it is a lossless process. You can encrypt a message and no data is lost. Retrieving the original message may be more or less difficult, depending upon whether or not you have the key to do it though.

 

We can of course, agree to disagree if the discussion appears argumentative to you.

 

-Paul

[sandyk]

This is your crazy opinion that has now become a legendary internet example of how nutty Computer Audiophiles can be. Please stop claiming it is substantiated fact, over and over again.

 

When hell freezes over, or I cark it, Mr. Know-it-all ill tempered Professor of another unrelated discipline .

You can't shut me up, because I have been able to demonstrate these things to many other people that are employed in these areas, as well as other Audiophiles.

Your childish polls won't silence me either. Grow up !

[Skeptic]

We must be talking about some different kind of interpolation.

 

Typically, one would up-sample the signal, and basically that is just stuffing zero value samples between the original samples. Adding more samples per time period is equivalent to increasing the sampling rate.

With PCM, you can add as many zeros as you like to increase the bit-depth from 16-bit to 24-bit, 32-bit, and beyond. This does nothing to change the original data at all, and you can "recover" the original 16-bit data by discarding them.

 

When you are upsampling the audio from say 44.1kHz to 88.2kHz you do not leave the original data intact and "fill in the gaps" (linear interpolation) or worse, simply duplicate the original sample points which will introduce severe aliasing as though it was a NOS DAC.

This is only going to make things sound worse.

 

Proper interpolation when upsampling is going to change all of the data and is not a losslessly reversible process.

You would be able to produce something which sounds similar to the original from that upsampled data, but it is not the original, and is degraded in quality from the original.

 

 

You should not be ripping to anything other than a losslessly compressed 16-bit 44.1kHz format with CD audio (typically FLAC or ALAC) or replacing those lossless files with converted ones.

[Paul R]

With PCM, you can add as many zeros as you like to increase the bit-depth from 16-bit to 24-bit, 32-bit, and beyond. This does nothing to change the original data at all, and you can "recover" the original 16-bit data by discarding them.

 

When you are upsampling the audio from say 44.1kHz to 88.2kHz you do not leave the original data intact and "fill in the gaps" (linear interpolation) or worse, simply duplicate the original sample points which will introduce severe aliasing as though it was a NOS DAC.

This is only going to make things sound worse.

 

Proper interpolation when upsampling is going to change all of the data and is not a losslessly reversible process.

You would be able to produce something which sounds similar to the original from that upsampled data, but it is not the original, and is degraded in quality from the original.

 

 

You should not be ripping to anything other than a losslessly compressed 16-bit 44.1kHz format with CD audio (typically FLAC or ALAC) or replacing those lossless files with converted ones.

 

You can up-sample and interpolate without increasing the bit depth.

 

Well, do you have any references to what you call "proper interpolation?" Because I sure don't know why any interpolation algorithm would destroy original audio data, and I am not at all sure most DSP interpolations are doing that.

 

Not challenging, just want to find out for myself here.

 

-Paul

[sandyk]

Paul

Have you also tried converting 24/96,or perhaps even 24/192 to DSF256 ?

Alex

[gmgraves]

Paul,

 

basically PCM -> DSD has to be lossy, as DSD -> PCM is lossy.

 

I'm sure we can agree on the truth of the statements A and B below.

 

A) Lossy is an antonym of lossless.

 

B) Lossless means that when converting from X to Y, all information from X can be recovered knowing only Y.

 

Obviously, PCM -> DSD is lossless if, and only if, DSD -> PCM is lossless.

 

 

I have a Sony XA777ES SACD player. It converts regular 16/44.1 CDs to DSD before decoding them to audio. It is without a doubt the best sounding CD player I've ever heard, and as a reviewer, I've had lot of CD decks (some very expensive) pass through my system. None sound as good as my Sony. So of LPCM to DSD is lossy, hurrah for lossy conversion!

[jhwalker]

I have a Sony XA777ES SACD player. It converts regular 16/44.1 CDs to DSD before decoding them to audio. It is without a doubt the best sounding CD player I've ever heard, and as a reviewer, I've had lot of CD decks (some very expensive) pass through my system. None sound as good as my Sony. So of LPCM to DSD is lossy, hurrah for lossy conversion!

 

No one is commenting on which sounds subjectively "better" to one listener or another, simply stating that *any* resampling or interpolation (with the exception of simply bit-padding from 16- to 24-bits, or the like) is destructive / lossy, as it is impossible to return to the original, unaltered bitstream.

[InfernoSTi]

It seems that PCM > DSD then DSD > Analog Tape then Analog Tape > PCM is not lossy as it avoids the dithering issues of direct conversion. Perhaps this is completely wrong. What about my thinking is wrong, please? I'm asking to learn...

 

John

[jhwalker]

It seems that PCM > DSD then DSD > Analog Tape then Analog Tape > PCM is not lossy as it avoids the dithering issues of direct conversion. Perhaps this is completely wrong. What about my thinking is wrong, please? I'm asking to learn...

 

John

 

Sounds like we're speaking a different language entirely. Every single step you list above is inherently lossy.

 

PCM > DSD = lossy. There is no direct transform that allows "conversion" between one and the other without losing the integrity of the original file.

 

DSD > analog (anything) = lossy. Digital > analog conversion is lossy, period. You cannot recover the original signal after conversion to analog.

 

Analog > PCM = lossy. You cannot convert analog signal to digital (any format) without simply approximating (digitizing) the analog signal.

 

The only reasonable definition for lossless conversion is between digital files (e.g., WAV, ALAC, FLAC, AIFF, etc.) in which the original content is 100% recoverable after conversion, bit perfect. Any other operation (upsampling, downsampling, conversion between formats, etc.) is lossy.

[kumakuma]

It seems that PCM > DSD then DSD > Analog Tape then Analog Tape > PCM is not lossy as it avoids the dithering issues of direct conversion. Perhaps this is completely wrong. What about my thinking is wrong, please? I'm asking to learn...

 

John

 

I think the preferred method is DSD to wax cylinder then back from wax cylinder to DSD. You need a linear power supply for the wax cylinder recorder though to avoid jitter.

[jhwalker]

I think the preferred method is DSD to wax cylinder then back from wax cylinder to DSD. You need a linear power supply for the wax cylinder recorder though to avoid jitter.

 

YES, back to basics is always best!

[sandyk]

You need a linear power supply for the wax cylinder recorder though to avoid jitter.

 

Did your wax cylinder recorder come with an SMPS, or do you wind up a clockwork motor ?

Perhaps you need to keep cranking the handle like an Organ Grinder and his monkey ?

Have you tried recording HDTV with it too ?

[kumakuma]

Did your wax cylinder recorder come with an SMPS, or do you wind up a clockwork motor ?

Perhaps you need to keep cranking the handle like an Organ Grinder and his monkey ?

Have you tried recording HDTV with it too ?

 

It actually came with a small monkey.

 

I waited until I found him a new home before replacing him. An added side benefit of replacing the power supply is the house smells fresher and the bite marks on my arms are healing up quite nicely.

 

I haven't tried HDTV yet. It sounds like an interesting summer project.

[Bob Stern]

DSD > PCM: Audiogate vs. Bitperfect DSD Master

 

If, for whatever reason, you want to convert DSD to PCM, the comparison tracks on the DSD Master website convince me that Bitperfect's DSD Master is better than Audiogate. In the final 5 minutes of Mahler 1, the timbres of the different instruments are much more differentiated in DSD Master than Audiogate.

 

DSD Master: What is DSD Master?

 

DSD Master: Download some Free Samples

[Paul R]

No one is commenting on which sounds subjectively "better" to one listener or another, simply stating that *any* resampling or interpolation (with the exception of simply bit-padding from 16- to 24-bits, or the like) is destructive / lossy, as it is impossible to return to the original, unaltered bitstream.

 

Wow - this discussion is going all over the place...

 

1.) PCM is not a bitstream, it is packet stream and requires decoding to a bitstream.

 

2.) Upsampling literraly is just stuffing zero packets between valid samples. It is no great trick to decimate the zero packets out of a file and return to the original file.

 

3.) I am still disputing that interpolation does not destroy the original data samples, and that at least in theory, those samples can be retrieved. By any definition, that makes it lossless.

 

I'm not so sure about the DSD conversion though, as that *is* a bitstream. I do believe that again at least in theory, you could reconstruct the original PCM data from it. But that is quite possibly wrong. I don't think that makes it a lossy conversion, as not data at all was "thrown away", no decimation occurred, and so forth. Just opinion on that though, I don't know what the formal definition of lossy is.

 

I do know if you can reconstruct the original file, then all the conversions involved had to be lossless.

 

The inverse of that is not true however, being unable to re-construct the original file does not automatically mean a lossy conversion took place. Just a conversion that is had to reverse.

 

-Paul

[InfernoSTi]

I learn something every day. What was lost? Hmm...not sound quality but something that one can't actually listen to...check sums!

 

It seems that "better" is no longer a synonym for "lossless" vs. "lossy". I will be careful when I say that my lossless file sounds better than my lossy file, especially if DSD files are "lossy" but sound better.

 

Best,

John