DeltaWave null-testing audio comparator (beta)

[esldude]

I've been busy and unable to try much the last two weeks.  You have added many neat new features and the precision improvement is welcome in the current version.  I like the Manual adjustments feature too.  I've managed to improve some results a couple db, but not more.  Optimize seems to get worse the more it runs so not sure what I think of it yet. 

 

I do have a question about the delta spectrogram.  You added the colored scale at the side at my request and it is a welcome addition.  The last three versions at least the delta spectrogram itself looks different than it used to look, and doesn't really make sense to me.  The scale chosen by the software seems much too narrow and it isn't showing anything of interest in the spectrogram.  The spectrograms for the files themselves still looks like you'd expect.  Here is one for Bob Marley reference vs 1st generation copy. 

 

Here is the delta spectrogram for the Marley reference vs 8th gen copy. 

 

The spectrogram of the compare file looks fine. 

[pkane2001]

1 hour ago, esldude said:

I've been busy and unable to try much the last two weeks.  You have added many neat new features and the precision improvement is welcome in the current version.  I like the Manual adjustments feature too.  I've managed to improve some results a couple db, but not more.  Optimize seems to get worse the more it runs so not sure what I think of it yet. 

 

I do have a question about the delta spectrogram.  You added the colored scale at the side at my request and it is a welcome addition.  The last three versions at least the delta spectrogram itself looks different than it used to look, and doesn't really make sense to me.  The scale chosen by the software seems much too narrow and it isn't showing anything of interest in the spectrogram.  The spectrograms for the files themselves still looks like you'd expect.  Here is one for Bob Marley reference vs 1st generation copy. 

 

Here is the delta spectrogram for the Marley reference vs 8th gen copy. 

 

The spectrogram of the compare file looks fine. 

 

Dennis, as you correctly noticed, the delta spectrogram has changed when I introduced the scale bar. Instead of showing the spectrum of the delta waveform, it is now showing the difference of spectra, instead. In other words, it's the over-time representation of the Δ of Spectra chart, instead of the Spectrum of Δ one.

 

The problem with the original delta spectrogram was that it was often nearly impossible to tell the difference between it and the original waveforms. 

[pkane2001]

 

2 hours ago, esldude said:

 I like the Manual adjustments feature too.  I've managed to improve some results a couple db, but not more.  Optimize seems to get worse the more it runs so not sure what I think of it yet. 

 

Optimize option may or may not work well. It will attempt to find a better combination of parameters, and in general will only show results that have at least one number that's an improvement over the previous results. Overall, that may not produce a good result, but in my testing, it often does suggest one or two reasonable changes to be looked at. Don't expect the improvement to occur at the end of the optimization run -- it could happen at the beginning or in the middle. It's just a search algorithm that doesn't understand the underlying parameters that it's trying to optimize. In effect it's a semi-intelligent trial-and-error. 

[esldude]

I don't know if this has always been there or added and I missed it.  On any of the charts, you can click and hold the mouse scroll wheel which then lets you draw a box around any part of the chart of interest.  When you release it the selected part of the chart expands to fill the chart area with scaling of the axis set appropriately.  I find this much quicker and more convenient than scrolling the chart or scrolling one axis of the chart.  I box the area of interest, and then hit reset axis when done.   Or sometimes I box part of the chart to zoom in, box a portion of that and zoom in further before resetting axis. 

 

EDIT to add:

You also can put the mouse pointer on either axis in the label area and do a scroll click to select a narrow section all the way across or all the way up.  

 

 

[esldude]

Looks like the new version has a Y-axis bug on any chart that displays a waveform. 

 

 

Also any method of zooming the y-axis causes it to change the labeled scale in reverse of what it should though the waveform display acts as expected.  

[pkane2001]

20 minutes ago, esldude said:

Looks like the new version has a Y-axis bug on any chart that displays a waveform. 

 

The 0 value at the center line is not in dB, as that can't be expressed in decibels (a log of 0 doesn't compute).

 

The scale is in dBFS, so it’s 0dB at 1.0 and -1.0 extremes and goes negative below that.

[esldude]

14 minutes ago, pkane2001 said:

 

The 0 value at the center line is not in dB, as that can't be expressed in decibels (a log of 0 doesn't compute).

 

The scale is in dBFS, so it’s 0dB at 1.0 and -1.0 and goes negative below that.

Shows I am still tired from working too much in the recent days.  I should have understood that.  Skipping some versions I didn't notice the db scale instead of the fraction of 1 used in previous versions.  So that was throwing me off.   Sorry for the false alarm.  

[pkane2001]

6 minutes ago, esldude said:

Shows I am still tired from working too much in the recent days.  I should have understood that.  Skipping some versions I didn't notice the db scale instead of the fraction of 1 used in previous versions.  So that was throwing me off.   Sorry for the false alarm.  

 

No problem! You can switch back to the fractional scale by unchecking the dB checkbox.

[fas42]

Nice work, Paul! ... An instance of where the precision is paying off, for that 0.5 sample discrepancy,

DeltaWave v1.0.22, 2019-04-01T08:33:03.5235453+11:00
Reference:  BM,44.1,orig.wav[L] 1574918 samples 44100Hz 32bits, stereo, MD5=00
Comparison: BM,44.1,orig,1s.wav[L] 1574918 samples 44100Hz 32bits, stereo, MD5=00
Settings:
    Gain:True, Remove DC:True
    Non-linear Gain:False    EQ FFT Size:128, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB
    Correct Drift:True, Precision:30
    Upsample:False, Window:Hann
    Spectrum Window:Blackman, Spectrum Size:524288
    Spectrogram Window:Lanczos, Spectrogram Size:32768, Spectrogram Steps:1024
    Dither:False
    Trim Silence:False

Discarding Reference:  Start=0s, End=0s
Discarding Comparison: Start=0s, End=0s

Initial peak values Reference: -1.548dB   Comparison: -1.49dB
Initial RMS values Reference: -17.93dB   Comparison: -17.93dB

Null Depth=49.125dB
X-Correlation offset: 0 samples

Final peak values Reference: -1.548dB   Comparison: -1.548dB
Final RMS values Reference: -17.93dB   Comparison: -17.93dB

Gain= 0dB (1x) DC=0 Phase offset=0.011338ms (0.5 samples)
Difference (rms) = -152.51dB [-155.86dBA]
Correlated Null Depth=162.41dB [164.33dBA]
Clock drift: 0 ppm

Files are NOT a bit-perfect match (match=99.96%) at 16 bits
Files are NOT a bit-perfect match (match=13.35%) at 32 bits
Files match @ 90.0329% when reduced to 24 bits

RMS of the difference of spectra: -210.047561472656dB
gn=1.00000000513874, dc=-1.05245235413156E-11, dr=0, of=0.5

DONE!

Signature: bfc2a6815be2dfa9ceef8a9c02f92636

[esldude]

Hopefully this isn't me just being tired and missing something obvious. 

 

When zooming in on a waveform for the top half above zero level I get this.  Notice the db level displayed with decimal places.  This works into even more decimal places than this example. 

 

But a similar level of zooming on the bottom half of the waveform results in this where the decimal places aren't shown.  You never get any db decimal places zooming on the bottom half of the waveform displayed. 

 

[pkane2001]

47 minutes ago, esldude said:

Hopefully this isn't me just being tired and missing something obvious. 

 

When zooming in on a waveform for the top half above zero level I get this.  Notice the db level displayed with decimal places.  This works into even more decimal places than this example. 

 

But a similar level of zooming on the bottom half of the waveform results in this where the decimal places aren't shown.  You never get any db decimal places zooming on the bottom half of the waveform displayed. 

 

That looks like I forgot to take an absolute value when computing scale. I’ll check and fix ASAP.

[pkane2001]

2 hours ago, esldude said:

I don't know if this has always been there or added and I missed it.  On any of the charts, you can click and hold the mouse scroll wheel which then lets you draw a box around any part of the chart of interest.  When you release it the selected part of the chart expands to fill the chart area with scaling of the axis set appropriately.  I find this much quicker and more convenient than scrolling the chart or scrolling one axis of the chart.  I box the area of interest, and then hit reset axis when done.   Or sometimes I box part of the chart to zoom in, box a portion of that and zoom in further before resetting axis. 

 

EDIT to add:

You also can put the mouse pointer on either axis in the label area and do a scroll click to select a narrow section all the way across or all the way up.  

 

 

 

To add to this: if you are zooming in/out and the chart seems to jump by a large factor with each scroll wheel adjustment, just hold down the Control key on the keyboard at the same time. The zoom will become much more fine-grained.

[fas42]

At the moment trying to null the original and 1st gen versions of Marley - still haven't got an audible difference better than about 40dB; that is, the delta wave sounds like a trebly version of the original, attenuated 40dB ... but working on it, .

 

A suggestion: the progress bar on the status line tells me nothing, and usually loses any relation to reality after some matches, barely showing any progression from the left, all through the processing. Would it be possible to show a steady display here of what is currently dumped into the log file, updating it as a single line as the processing proceeds through the steps?

[pkane2001]

1 hour ago, fas42 said:

A suggestion: the progress bar on the status line tells me nothing, and usually loses any relation to reality after some matches, barely showing any progression from the left, all through the processing. Would it be possible to show a steady display here of what is currently dumped into the log file, updating it as a single line as the processing proceeds through the steps?

 

I'll fix the progress bar. Didn't realize it was broken

[fas42]

Got this result for a short period comparison,

 

Null Depth=6.826dB
X-Correlation offset: 1 samples
Stopped! All sample vectors must have the same length. However, vectors with disagreeing length 417792 and 387071 have been provided. A sample with index i is given by the value at index i of each provided vector.
Signature: f1fe33c68b85b8a8ae858eec9317b068

 

Also, a big plea, again, for being able to lock the axes of other plots when zooming - thanks!

[fas42]

Trying some experiments with shorter segments. and manual adjustment - added EQ, which made it worse; an extreme example, 

 

DeltaWave v1.0.22, 2019-04-01T15:11:50.8024956+11:00
Reference:  Bob Marley original.wav[L] 608256 samples 44100Hz 32bits, stereo, MD5=00
Comparison: Bob Marley 1st gen.wav[L] 692224 samples 44100Hz 32bits, stereo, MD5=00
Settings:
    Gain:True, Remove DC:True
    Non-linear Gain:True    EQ FFT Size:524288, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB
    Correct Drift:True, Precision:30
    Upsample:False, Window:Hann
    Spectrum Window:Blackman, Spectrum Size:524288
    Spectrogram Window:Lanczos, Spectrogram Size:32768, Spectrogram Steps:1024
    Dither:False
    Trim Silence:False

Discarding Reference:  Start=0s, End=22s
Discarding Comparison: Start=1.2s, End=20.8s

Initial peak values Reference: -2.011dB   Comparison: -1.94dB
Initial RMS values Reference: -18.131dB   Comparison: -18.197dB

Null Depth=8.359dB
X-Correlation offset: -2543 samples

Final peak values Reference: -2.011dB   Comparison: -1.881dB
Final RMS values Reference: -18.131dB   Comparison: -18.043dB

Gain= -0.9284dB (0.8986x) DC=-0.00001 Phase offset=-57.677086ms (-2543.56 samples)
Difference (rms) = -33.45dB [-38.6dBA]
Correlated Null Depth=34.32dB [27.99dBA]
Clock drift: 0.34 ppm

Files are NOT a bit-perfect match (match=0.05%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 49.9985% when reduced to 5.69 bits

RMS of the difference of spectra: -116.344440993497dB
gn=1.11280373966288, dc=-6.66333176665983E-06, dr=3.368E-07, of=-2543.5595

DONE!

Signature: 4bb29be426b5437a7f3c2a5e993655c8

 

 

 

[pkane2001]

8 hours ago, fas42 said:

Got this result for a short period comparison,

 

Null Depth=6.826dB
X-Correlation offset: 1 samples
Stopped! All sample vectors must have the same length. However, vectors with disagreeing length 417792 and 387071 have been provided. A sample with index i is given by the value at index i of each provided vector.
Signature: f1fe33c68b85b8a8ae858eec9317b068

 

That error happens if you have less than 10 seconds worth of samples in one or both of the tracks. I can change it so that it works with fewer samples, but accuracy will be lowered.

 

Quote

Also, a big plea, again, for being able to lock the axes of other plots when zooming - thanks!

In other words, you want to allow the current plot to be zoomed in/out without affecting the scale of the other plots, correct?

[TomCapraro]

42 minutes ago, pkane2001 said:

 

That error happens if you have less than 10 seconds worth of samples in one or both of the tracks. I can change it so that it works with fewer samples, but accuracy will be lowered.

 

In other words, you want to allow the current plot to be zoomed in/out without affecting the scale of the other plots, correct?

No, please ... now the software is very precise, I won't reduce the accuracy otherwise I won't download a new version 😂

[TomCapraro]

If there is something to be corrected in the dB scale it is fine but I would advise not to change the alignment accuracy, now it's very good.

[pkane2001]

9 minutes ago, TomCapraro said:

No, please ... now the software is very precise, I won't reduce the accuracy otherwise I won't download a new version 😂

 

Of course I'm not going to do something that will lower accuracy! This was just a suggestion to allow shorter audio clips to work without an error. But, shorter clips will always produce a lower accuracy of alignment or the match operation will fail. That's just how statistics work in the presence of noise.

[Arpiben]

 

Out of concern but by curiosity, which type of Blackman Harris FFT window are you using? 3- 4  terms one ?

I do appreciate the availability of all spectra views since they allow  fast detection of buried details.

DW v1.0.22b:

Phase unwrapping is working as expected 😊 

Lissajous cursor is still disappearing in certain extreme  circumstances (768kSamples * 30s) but I can manage. 😉

 

Sincerely you have been doing a very great job 👍

[pkane2001]

40 minutes ago, Arpiben said:

 

Out of concern but by curiosity, which type of Blackman Harris FFT window are you using? 3- 4  terms one ?

 

This one: https://en.wikipedia.org/wiki/Window_function#Blackman–Harris_window

 

[Arpiben]

44 minutes ago, pkane2001 said:

 

This one: https://en.wikipedia.org/wiki/Window_function#Blackman–Harris_window

 

 

I was expecting one of those but it doesn't matter since it is possible to retrieve which one DW is using. 

7-term is having lower side lobes at the expense of a wider main lobe.

 

 

[pkane2001]

7 minutes ago, Arpiben said:

 

I was expecting one of those but it doesn't matter since it is possible to retrieve which one DW is using. 

7-term is having lower side lobes at the expense of a wider main lobe.

 

 

 

I can always add a few more windows to see if they work any better

[fas42]

8 hours ago, pkane2001 said:

 

That error happens if you have less than 10 seconds worth of samples in one or both of the tracks. I can change it so that it works with fewer samples, but accuracy will be lowered.

 

I would suggest allowing it, with a warning that accuracy may be compromised, in the report.

 

8 hours ago, pkane2001 said:

 

In other words, you want to allow the current plot to be zoomed in/out without affecting the scale of the other plots, correct?

 

Yes. And make it easily switchable, on or off; a checkbox, say.