DeltaWave null-testing audio comparator (beta)

[esldude]

54 minutes ago, fas42 said:

 

No dither, ever. I always work at 32 bit floating, and only when exporting to a WAV file do I select the required format. I have done experiments in Audacity a number of times of upsampling and downsampling to an extreme degree, and then diff'ing after going through a number of cycles - the accuracy is excellent, way below -90dB levels, say.

 

If you upload the first generation, that would be excellent, thanks!

That brings up a point @pkane2001 might know about or want to know.  I need to redo it.  But I was saving some files without dither in Audacity.  Just to convert from either 16 or 24 bit to 32 bit float. When you save them to 32 bit floating it seems the nulls and matching were better no matter what the reference file in use was.  I believe he has said Deltawave works internally all 32 bit so I wonder why saving a file to it helps?  Again I need to repeat this and make sure I didn't get it mixed up along the way.  

[fas42]

Just to mention a couple of requests: could the zoom in the time axis have meaningful numbers when you go right in, and can the trim at the front (and end) allow high precision, in both secs and no. of samples?

[fas42]

And I have just noticed, , that the labelling of the time axis, is not consistent - 5.73, 5.74, 5.76. 5.77! Can anything be done about that, or is it intrinsic to the plotting package?

[esldude]

@fas42

 

Okay here it is>

 

https://www.dropbox.com/s/0jciymo33xv14sx/Bob Marley all generations.zip?dl=0

 

I went ahead and saved it all in 32 bit float to save you a bit of trouble.  I also included the 1st and 4th gen in 24 bit, and the original reference in its original 16 bit form as well as 32 float for all of these.  Unless the file title says otherwise these are in 32 bit float. 

 

I was not particularly careful about trimming the ends.  You may wish to even that up or you can see how Deltawave works with it. 

[fas42]

Beautiful !! ... Thanks!

[Arpiben]

9 hours ago, pkane2001 said:

 

Thanks for the log -- I'll review ASAP.

 

Axis reset to 22KHz is a leftover from the days when the plots were painfully slow to render. I had to rewrite large parts of the plot library to speed this up. Back then, I tried to limit the initial display to just the audible range. I'll fix this, but for now, it's easy enough to zoom out to the full range.

 

As far as chirps are concerned, that's a known problem. Something I've been discussing with @esldude recently. The issue is that a sine chirp consists of a repeated waveform. While the amplitude is changing, the frequency remains the same. For a phase-driven alignment algorithm there is nothing to distinguish between one period of a sine wave and another. All of them produce the same perfect match, and this is where things fall apart. There are too many possible solutions. I do intend to come up with a solution to this, but for now, 'naturally occurring' waveforms will produce a much better result than anything generated

 

 

 

No worries.

All started with why correct clock drifts with non linear interpolation taking into account how oscillators behave?

Then I wanted to evaluate the linear correction applied....I guess I will need to change method 😉

 

[pkane2001]

9 hours ago, esldude said:

That brings up a point @pkane2001 might know about or want to know.  I need to redo it.  But I was saving some files without dither in Audacity.  Just to convert from either 16 or 24 bit to 32 bit float. When you save them to 32 bit floating it seems the nulls and matching were better no matter what the reference file in use was.  I believe he has said Deltawave works internally all 32 bit so I wonder why saving a file to it helps?  Again I need to repeat this and make sure I didn't get it mixed up along the way.  

 

 

DeltaWave is using double floating point format internally, 64 bits samples.

 

So I tried an experiment. First, I took the Bob Marley A track and loaded it into Audacity, saved as 16 bit WAV file. Loaded the original and the 16-bit file into DeltaWave. The result? Obviously something is happening at the high-frequency end of the spectrum. The files become a bit-perfect match at about 13 bits, so we changed up to about 2-3 lower bits in this process.

 

I then saved the same file from Audacity as 32-bit floating point, and compared to the original. The result? Bit perfect at 100%!

 

Then, went into Audacity and set the following quality settings to turn of all possibility of dither:

 

Then exported the same file as 16-bit WAV.. Result? 100% Bit perfect.

 

So it looks like you just need to be very careful about the settings in Audacity if you want unmodified data to be exported.

 

[mansr]

13 minutes ago, pkane2001 said:

So it looks like you just need to be very careful about the settings in Audacity if you want unmodified data to be exported.

Yes, Audacity is very eager to apply its shaped dither to anything it touches.

[pkane2001]

22 minutes ago, mansr said:

Yes, Audacity is very eager to apply its shaped dither to anything it touches.

 

Here's the difference in spectrum that shaped dither causes on this track:

 

 

[esldude]

I apparently got confused along the way.  I re-did some of these and they work out the same.  I'm aware of Audacity's dither and have it turned off unless doing rate or bit depth conversions.  Had it off when saving to 32 bit float.  So float works the same as regular PCM on the same file like it should.  Sorry for the false concern. 

[fas42]

Having major issues still with Internet; currently gone bad again. So no pics... 

 

Looked at 1st gen copy of Marley, and in that 1.749 centred sample I posted above, the damage is already starting to happen - cumulative, as one would expect. 

[fas42]

Something that could be very, very interesting - looked at the 4th, and it's almost the same as the 8th! So, implying that only so much distortion occurs; there's a limit... Will need to carefully check this out... 

[pkane2001]

1 hour ago, fas42 said:

Something that could be very, very interesting - looked at the 4th, and it's almost the same as the 8th! So, implying that only so much distortion occurs; there's a limit... Will need to carefully check this out... 

 

Keep in mind that whatever filter the DAC and ADC are using in the process will have the most effect on the first pass. Once the damage is done, for the rest of the passes the effect will be a smaller, and so the subsequent delta will be less. Try low-pass filtering (in DW) just around where the filter starts to kick in (for Dennis' files that's a bit above 19kHz) and see if that helps paint a clearer picture.

 

[fas42]

This trend is strong... Obvious thing was to compare 4th and 8th - - and this confirmed a very good to excellent match - to as far left or right as I wanted to go. 

 

Further confirmation... Looked at 6th gen, compared to 4th - superb match... 

 

My estimate is that the 3rd is the end of distorting cycle - the circuitry can't process the input waveform; and "smooths" it until it can... Will examine the 3rd next... 

[fas42]

From Paul's comment, will look closely at the waveform, in Audacity. 

[pkane2001]

I'm getting close to completing all the planned features for v1 of DeltaWave.... but not yet 

Version 1.0.18 is now available, and has the following changes/features:

Two new blind/sighted tests added, in addition to the ABX test that was already available:

 

• Added Stereo XY difference test to the Comparator. This is a pair-wise discrimination test that lets you use stereo perception to detect differences between two similar tracks
• 

  

• Added Preference XY difference test to the Comparator. This test lets you determine if you really like the sound of one track over another. As such, this is a subjective blind preference test
• 
   
• Added file and settings information to the test result window for validation, along with signature and file MD5 hashes

Other changes:

• Changed the behavior of play buttons and menu items, including the comparator to not require a Show or Match button press
• Extended default frequency scale on plots to Nyquist frequency

The two new test formats are experimental. I welcome all input as to whether they work, make sense, or maybe are completely misguided.  
@Jud

[fas42]

Getting results in Audacity which are not making sense, so far. I Brickwalled at 19kHz, and the waveform is showing frequency changes - is this Audacity "misbehaving"? 

 

 

[Jud]

1 hour ago, pkane2001 said:

I'm getting close to completing all the planned features for v1 of DeltaWave.... but not yet 

Version 1.0.18 is now available, and has the following changes/features:

Two new blind/sighted tests added, in addition to the ABX test that was already available:

 

• Added Stereo XY difference test to the Comparator. This is a pair-wise discrimination test that lets you use stereo perception to detect differences between two similar tracks
• 

  

• Added Preference XY difference test to the Comparator. This test lets you determine if you really like the sound of one track over another. As such, this is a subjective blind preference test
• 
   
• Added file and settings information to the test result window for validation, along with signature and file MD5 hashes

Other changes:

• Changed the behavior of play buttons and menu items, including the comparator to not require a Show or Match button press
• Extended default frequency scale on plots to Nyquist frequency

The two new test formats are experimental. I welcome all input as to whether they work, make sense, or maybe are completely misguided.  
@Jud

 

I like not requiring Show or Match.  Some of the mouseover information in the Stereo Comparator was a little confusing to me in a rushed two minute tryout this evening.  I'll see whether it becomes clearer tomorrow.  

 

What I ultimately want to do is this: Save a file doing a stereo comparison where the reference track is the left channel; and save a file doing a stereo comparison where the reference track is the right channel.  This will enable some MQA vs. hi res comparisons I want to do using DeltaWave with Audirvana+ and/or HQPlayer.

 

 

[pkane2001]

23 minutes ago, fas42 said:

Getting results in Audacity which are not making sense, so far. I Brickwalled at 19kHz, and the waveform is showing frequency changes - is this Audacity "misbehaving"? 

 

Try applying a 19k low pass filter in DeltaWave. If you compare the original to the 19k brickwalled version directly, the null will not be as good due to differences above 19k.

 

 

I get about a 92dB null, which is less than a 1 bit difference for a 16bit file. Don't see any abnormal new frequencies. Delta of spectra, zoomed-in (see the left scale!)

 

[pkane2001]

26 minutes ago, Jud said:

 

I like not requiring Show or Match.  Some of the mouseover information in the Stereo Comparator was a little confusing to me in a rushed two minute tryout this evening.  I'll see whether it becomes clearer tomorrow.  

 

What I ultimately want to do is this: Save a file doing a stereo comparison where the reference track is the left channel; and save a file doing a stereo comparison where the reference track is the right channel.  This will enable some MQA vs. hi res comparisons I want to do using DeltaWave with Audirvana+ and/or HQPlayer.

 

Don't know if I have the best descriptions in the tooltips, but the website has a bit more information on this (and other) tests:

 

https://deltaw.org/comparator.html

 

Look for the X/Y Stereo Difference Test section for instructions.

 

[fas42]

Purely done in Audacity, this is what the amplitude envelope looks like,

 

 

Zooming in,

 

 

and again,

 

 

No aspect of DW contaminated this story, - what gives??

[STC]

7 hours ago, fas42 said:

My estimate is that the 3rd is the end of distorting cycle - the circuitry can't process the input waveform; and "smooths" it until it can... Will examine the 3rd next... 

 

Are you suggesting that after the third generation the files will no longer degrade?

[fas42]

3 hours ago, STC said:

 

Are you suggesting that after the third generation the files will no longer degrade?

 

It's still unclear to me what's happening ... the copied sample alters for the first 3 passes, and then stabilises. Is this an artifact of the filtering of the converter circuitry as Paul suggests, and if so, should it be considered to be reasonable behaviour - or a significant distortion?

 

What I'm certain of is that I hear the difference between the original, and the 8th copy - on my laptop speakers. So there is something in the waveform that matters, and I'm aiming to track it down ... at the moment I'm looking for clues, via DW - it may turn out that what I mentioned in recent posts is in fact a red herring, in terms of being part of the relevant difference.

[Jud]

8 hours ago, pkane2001 said:

 

Don't know if I have the best descriptions in the tooltips, but the website has a bit more information on this (and other) tests:

 

https://deltaw.org/comparator.html

 

Look for the X/Y Stereo Difference Test section for instructions.

 

 

Got it, thanks.  I'm pretty amazed at the pace at which you're making changes in response to requests, and grateful.  

 

What remains in order to do the comparisons I would like, as mentioned in my previous note, is only one thing not currently present in DeltaWave, and that is to enable the mixed down Reference track to be compared and saved in either the right channel or the left.  

 

One reason, as I'm sure you already understand, is to be able to determine whether any preference results from differences between left and right ears, differences in speaker-room interactions between left and right, or is due to differences in the tracks themselves.  If preference switches depending on which channel the reference is played from, it's the track; if it doesn't switch, it's ears or speakers.

 

Saving a file with the comparison, which DeltaWave already has (though not with the Reference file in the right channel as far as I can tell) allows the application of various Audirvana+ and HQPlayer upsampling filters on playback.  For me and many others who use the streaming capabilities of these players, it also allows playback through our main systems.

[STC]

37 minutes ago, fas42 said:

the copied sample alters for the first 3 passes, and then stabilises.

 

I wonder how many passes the so called original copy went thru before reaching Esldude. So why an additional of three needed to stabilize?  Maybe something else you are overlooking.