DeltaWave null-testing audio comparator (beta)

[fas42]

3 hours ago, pkane2001 said:

 

Thanks for testing, Frank! Let's walk through your tests, to make sure I got everything right.

 

1. You compared Crowd Chant track A and B without doing any filtering on them. This produced a delta waveform with peaks of 1/2 of the original track, correct? What did you adjust that made it better? Offset, gain, drift?

 

As implied by my post, simply altered the offset value by 2 samples, integer 2 change, which brought them in very good alignment.

 

3 hours ago, pkane2001 said:

2. When you band-pass filter 1k-8k range, you are obviously removing some important data. But, it may just be that the warning about using alternate method was a bit too aggressive.  I also got the same warning, but when I said 'No' to the prompt to use the alternative method, here's what I got:

 

Seems like a pretty good result. I'll review what threshold to use for when this warning message pops up, as the logic was based on the poor precision of the offset calculation in previous versions. Can you please try answering 'No' to the alternative drift correction method to see if you get an improved result?

 

 

Haven't tried that, will do so ...

[fas42]

Another quick question - any reason for the set of values in the filter frequency selection dropdown? Could at least every octave value be present, e.g. 1000, 2000, 4000, 8000.

[fas42]

The mind's a funny thing ... even though the No option was available, it didn't trigger that this meant that I could continue with the first analysis findings - which, selecting No this time, gave me a nulling minimum part way through the result, partially countering the poor offset.

[pkane2001]

27 minutes ago, fas42 said:

Another quick question - any reason for the set of values in the filter frequency selection dropdown? Could at least every octave value be present, e.g. 1000, 2000, 4000, 8000.

 

No good reason, just some values I found useful during testing 

 

[pkane2001]

32 minutes ago, fas42 said:

The mind's a funny thing ... even though the No option was available, it didn't trigger that this meant that I could continue with the first analysis findings - which, selecting No this time, gave me a nulling minimum part way through the result, partially countering the poor offset.

 

My mistake. The condition that triggered the 'alternative' warning message was reversed, so that higher sampling rates required much greater precision in the offset calculation than lower rate ones. That's the opposite of what I intended to do 

[pkane2001]

DeltaWave version 1.0.27b is now available

 

The following changes are in the new version:

• Added upsample/downsample rate selection in settings
• Pressing play button next to any other file while one is playing will start playing the new file where the previous one left off
• Improved stability of FFT calculations to eliminate tiny differences in the least significant digits of double-precision numbers
• Added a list of 20 recently used files to the File menu
• Changed the unlock axis button to remember a separate zoom level for each of the charts
• Fixed the issue with chart zoom reset when using the Manual Adjustments tool
• Fixed offset value set to zero when using Cached data for the first time in Manual Adjustment tool
• Improved the accuracy of cross-correlation computation in the presence of a large clock drift
• Fixed the condition for prompting to use alternate drift correction algorithm when using higher resolution sampling rates

All issues, suggestions and comments are welcome, as always!

[esldude]

The software is really good now.  And the list of useful features has become extensive.  If we don't find any gotchas, are you about ready to have a non-beta version?

 

I'll have to test out the new large clock drift capability.  

[TomCapraro]

The previous version perfectly recovered the clock drift on two signals (same DAC) acquired from spdif input and usb input.
This version provides much worse results.😞

[pkane2001]

2 minutes ago, TomCapraro said:

The previous version perfectly recovered the clock drift on two signals (same DAC) acquired from spdif input and usb input.
This version provides much worse results.😞

 

Hi Tom,

 

Strange, can you share the files that generated this?

 

 

[pkane2001]

4 hours ago, esldude said:

The software is really good now.  And the list of useful features has become extensive.  If we don't find any gotchas, are you about ready to have a non-beta version?

 

I'll have to test out the new large clock drift capability.  

 

Pretty much so, Dennis. There were many internal computation changes in this last version. Once I address any issues reported in it, I’m ready to call the end to the beta test

 

[TomCapraro]

11 minutes ago, pkane2001 said:

 

Hi Tom,

 

Strange, can you share the files that generated this?

 

 

Yes 

to better understand, try 1.0.26
Even the null test of the DAC-ADC loopback is worse than version 1.0.26

 

http://www.filedropper.com/nuovacartella

[Jud]

5 minutes ago, pkane2001 said:

 

Pretty much so, Dennis. There were many internal computation changes in this last version. Once I address any issues reported in it, I’m ready to call the end to the beta test

 

 

Apologies for not having followed this particular point, but does the software have the capability now of using either channel as the "original"?  To me it may give one channel a slight advantage in simultaneous comparisons if it is always the template to which the selection on the other channel is adjusted.  (Sorry if I'm not expressing this well. It's the first time I've thought about it in a while and I don't recall all the details.)

[pkane2001]

18 minutes ago, Jud said:

 

Apologies for not having followed this particular point, but does the software have the capability now of using either channel as the "original"?  To me it may give one channel a slight advantage in simultaneous comparisons if it is always the template to which the selection on the other channel is adjusted.  (Sorry if I'm not expressing this well. It's the first time I've thought about it in a while and I don't recall all the details.)

 

Sure, Jud. You can chose the desired channel you want to compare separately from each file. You can also use their combination, L+R, if you'd like.

[pkane2001]

54 minutes ago, TomCapraro said:

Yes 

to better understand, try 1.0.26
Even the null test of the DAC-ADC loopback is worse than version 1.0.26

 

http://www.filedropper.com/nuovacartella

 

Thanks, Tom. This seems to have to do with the initial offset difference in the two files, so likely due to the cross-correlation change I made. Let me see if I can make it less sensitive to the initial misalignment in the two files. If you skip 1 second in the SPDIF file, the match becomes much better:

 

 

[fas42]

The changes have made the software much zippier in use, I likee! First funny I've come across ... I regularly use the custom zoom level to visually find a more precise value of say the x axis shift between reference and comparison, and suddenly the zoom level button stopped working in the Matched panel - still worked in other panels, but not in the one I wanted it to!

 

Ah-hah! I've got it; the button doesn't respond when the axes are unlocked.

[TomCapraro]

8 hours ago, pkane2001 said:

 

Thanks, Tom. This seems to have to do with the initial offset difference in the two files, so likely due to the cross-correlation change I made. Let me see if I can make it less sensitive to the initial misalignment in the two files. If you skip 1 second in the SPDIF file, the match becomes much better:

 

 

If I do the null test with the old version and save the DELTA file, loading it on the new version generates a graphic on the FFT with + 42dB of difference.

[fas42]

If an upsampling rate is selected, this is incompatible in the Manual Corrections window with Cache File Data selected - matching is completely screwed up.

[pkane2001]

10 hours ago, fas42 said:

If an upsampling rate is selected, this is incompatible in the Manual Corrections window with Cache File Data selected - matching is completely screwed up.

 

Hmm. Not sure why that would be. The upsampling operation is independent of whether something is cached or not. Let me test when I get back to my computer later this weekend.

 

[pkane2001]

17 hours ago, TomCapraro said:

If I do the null test with the old version and save the DELTA file, loading it on the new version generates a graphic on the FFT with + 42dB of difference.

 

As I said before, I think this has to do with trimming initial mis-matched portions of the waveforms. The reason is that I'm using a smaller portion of the files to compute the initial offset than in the previous version (the cross-correlation step). If the initial offset is off by more than a second or two, this is what causes the issue. Previous versions could work with initial mismatch of about 5 seconds.

 

I'll fix this ASAP.

[fas42]

Just came across something when trying a comparison, 100Hz HP@start, of some of those Gearslutz samples, no preliminary resampling - the filtering and matching is enough to cause clipping, which could be an issue when listening to playback - no problems on a full bandwidth comparison,

 

 

Perhaps a final check, and gain adjustment if necessary?

[pkane2001]

On 5/10/2019 at 7:45 PM, fas42 said:

Just came across something when trying a comparison, 100Hz HP@start, of some of those Gearslutz samples, no preliminary resampling - the filtering and matching is enough to cause clipping, which could be an issue when listening to playback - no problems on a full bandwidth comparison,

 

 

Perhaps a final check, and gain adjustment if necessary?

 

Ultimately, DW tries to match up gain levels using simple arithmetic, an offset and a multiplier. If that results in clipping at some points in the waveform, I'm not sure there's anything to be done. Internally, clipping is not an issue for DW computations, since everything is a floating point number, and can be much greater than 1 without causing distortions. When playing back, this could result in audible clipping. I can add an indicator showing that clipping has occurred. Would that be enough?

 

[fas42]

34 minutes ago, pkane2001 said:

 

Ultimately, DW tries to match up gain levels using simple arithmetic, an offset and a multiplier. If that results in clipping at some points in the waveform, I'm not sure there's anything to be done. Internally, clipping is not an issue for DW computations, since everything is a floating point number, and can be much greater than 1 without causing distortions. When playing back, this could result in audible clipping. I can add an indicator showing that clipping has occurred. Would that be enough?

 

 

Since depending on the playback setup glitches could be audible, to me it makes sense that there is at least an option to adjust levels of the matched waveforms, and delta if someone is so picky to worry about that, so that there is some slight headroom. Perhaps, prompt the user that there is clipping, and let him decide whether to leave as is, or request simple gain adjustment to be applied - obviously, once prompted for a certain pair, and settings, don't repeat prompt until something changes.

[pkane2001]

DeltaWave 1.0.28 is now available with some fixes and enhancements:

• Fixed trim calculation to allow at least 5 seconds of mismatch at the file start (previous version reduced this to 1 sec)
• Fixed resampling option from Manual Adjustment window when caching is enabled
• Added the number of clipped samples indicator and Correct Clipping process menu option to rescale to 0dBFS or less

@TomCapraro, please check if this fixes the problem with your USB/SPDIF tests.

[TomCapraro]

I think the problem is not just about alignment, I extracted the delta (usb / spdif) this if I load it on version 1.0.26 it gives me a FFT 65536 points at almost -130dB minimum, while if I load the delta on version 1.0 .28 the FFT gives me a higher value of at least 43dB.
I posted you the delta so you can try and understand better.
There is a problem with the FFT display.

http://www.filedropper.com/nuovacartella_1

[esldude]

I'm seeing the same problem as Tom. 

 

I compared one set of interconnects vs another.  Difference rms was -98 db.  In .26 this displays using a 2 k FFT as mostly lying between -115 and -120.  Doing this in .28 I get pretty much identical results on all the parameters.  But displaying the spectrum of delta now shows most of the spectrum around -72 db.  Like Tom an over 40 db difference.  And clearly too high for the results.  

 

So it is working, but the displayed FFT has a problem somewhere.  Results are fine, other displays are fine, but the spectrum of delta isn't right.