DeltaWave null-testing audio comparator (beta)

[Arpiben]

On 10/17/2019 at 3:53 AM, pkane2001 said:

 

In case you want to do some really extreme testing with a 64-bit version of the multitone file, I generated one to see what that would look like. The phases are not exactly as produced by AP but the frequencies are. I didn't try to guess the algorithm they used for phase. Instead, I just used a common one (Shapiro-Rudin).

 

Here's a comparison between the AP 32-bit and my 64-bit files. I dithered the 64-bit file at 70 bits  something you can do with floating point numbers. Of course, the AP file is in blue:

 

Hi Paul,

 

I have been generating ( for fun and learning purposes) multi tones with low crest factor ( Golay_Rudin_Shapiro, Newman, Schroeder, Kitayoshi, etc...). The point is I didn't succeed yet retrieving the algorithm for matching the frequencies of AP file. In principle, it should not be so difficult but I am struggling... Do you mind sharing how did you generate them?

Here are some phases :

 

Newman:

Schroeder:

Kitayoshi:

[TomCapraro]

33 minutes ago, Arpiben said:

@TomCapraro

 

Please, can you share your spectrum settings since I am also having correct harmonic levels with Kaiser & Dirichlet?

Rgds

 

[TomCapraro]

 

This is instead the report with Kaiser setting

 

 

Comparison THD   = -28,26dB
    H1 (1000Hz) = 0,32dB
    H3 (3000Hz) = -27,07dB
    H5 (5000Hz) = -40,77dB
    H7 (7000Hz) = -68,04dB
    H9 (9000Hz) = -85,74dB

[Arpiben]

16 minutes ago, TomCapraro said:

 

 

Now we are phase aligned 😊. I am getting your results.

 

Comparison THD   = -26,07dB
    H1 (1000Hz) = -1,78dB
    H3 (3000Hz) = -102,16dB
    H5 (5000Hz) = -70,5dB
    H7 (7000Hz) = -94,43dB

[TomCapraro]

10 minutes ago, Arpiben said:

 

Now we are phase aligned 😊. I am getting your results.

 

Comparison THD   = -26,07dB
    H1 (1000Hz) = -1,78dB
    H3 (3000Hz) = -102,16dB
    H5 (5000Hz) = -70,5dB
    H7 (7000Hz) = -94,43dB

Aren't they weird?

[Arpiben]

6 minutes ago, TomCapraro said:

Aren't they weird?

 

Yes from my point of view, but I am not THD expert. It seems that algorithm calculates the difference between Ref&Comp harmonic at first glance. But there is probably some power density calculation within a specified bandwidth which might be tricked...No idea. Paul will help us 😉

[esldude]

Okay, using Blackman Harris 7 I get this.

 

Comparison DR = 52.2dB

Comparison THD+N = -27.32dB

Comparison THD   = -28.44dB
    H1 (1000Hz) = -2.73dB
    H3 (3000Hz) = -32.77dB
    H5 (5000Hz) = -52.18dB
    H7 (7000Hz) = -89.33dB
    H9 (9000Hz) = -118.18dB

 

Using Hann the 3rd harmonic level is clearly wrong. 

Comparison DR = 48.55dB

Comparison THD+N = -27.32dB

Comparison THD   = -26.06dB
    H1 (1000Hz) = -4.5dB
    H3 (3000Hz) = -105.85dB
    H5 (5000Hz) = -73.23dB
    H7 (7000Hz) = -97.13dB

 

Using Dirichlet no H3 is listed.  There is one there in the spectrum at about -27 db. 

Comparison DR = 31.52dB

Comparison THD+N = -27.32dB

Comparison THD   = -41.64dB
    H1 (1000Hz) = -9.61dB
    H5 (5000Hz) = -53.17dB
    H7 (7000Hz) =  -71dB

 

This is Nutall:

Comparison DR = 53.53dB

Comparison THD+N = -27.32dB

Comparison THD   = -26.69dB
    H1 (1000Hz) = -3.39dB
    H3 (3000Hz) = -39.42dB
    H5 (5000Hz) = -77.8dB

 

This is Kaiser:

Comparison DR = 51.07dB

Comparison THD+N = -27.32dB

Comparison THD   = -28.26dB
    H1 (1000Hz) = -2.41dB
    H3 (3000Hz) = -29.8dB
    H5 (5000Hz) = -43.5dB
    H7 (7000Hz) = -70.78dB
    H9 (9000Hz) = -88.49dB

 

Here is Flat top windowing. 

Comparison DR = 50.84dB

Comparison THD+N = -27.32dB

Comparison THD   = -26.73dB
    H1 (1000Hz) =   -2dB
    H3 (3000Hz) = -29.37dB
    H5 (5000Hz) = -51.41dB
    H7 (7000Hz) = -106.46dB

 

ON music I typically use blackman-harris 7 or Hann.  

 

 

[pkane2001]

11 hours ago, Arpiben said:

 

Hi Paul,

 

I have been generating ( for fun and learning purposes) multi tones with low crest factor ( Golay_Rudin_Shapiro, Newman, Schroeder, Kitayoshi, etc...). The point is I didn't succeed yet retrieving the algorithm for matching the frequencies of AP file. In principle, it should not be so difficult but I am struggling... Do you mind sharing how did you generate them?

Here are some phases :

 

Newman:

Schroeder:

Kitayoshi:

 

Hi Arpiben,

 

As I mentioned, I didn't succeed (or tried) to match the phase of the AP file. I was primarily interested in getting a reasonably low crest multi-tone sample generated in 64 bits instead of the AP 32. The algorithm I used was Shapiro-Rudin. 

 

I've been thinking about adding a Generator option to DW for a while to create various test tones, but thought it would be more of a feature just for me and not that useful to anyone else  Maybe I'll do that anyway, since I have the code already written, but only used during testing.

 

[Arpiben]

3 minutes ago, pkane2001 said:

 

Hi Arpiben,

 

As I mentioned, I didn't succeed (or tried) to match the phase of the AP file. I was primarily interested in getting a reasonably low crest multi-tone sample generated in 64 bits instead of the AP 32. The algorithm I used was Shapiro-Rudin.

 

Hi Paul,

 

But at least you got the frequencies right , not yet my case. AP's frequencies aren't equally spaced.😉 

[pkane2001]

3 hours ago, esldude said:

Okay, using Blackman Harris 7 I get this.

 

Comparison DR = 52.2dB

Comparison THD+N = -27.32dB

Comparison THD   = -28.44dB
    H1 (1000Hz) = -2.73dB
    H3 (3000Hz) = -32.77dB
    H5 (5000Hz) = -52.18dB
    H7 (7000Hz) = -89.33dB
    H9 (9000Hz) = -118.18dB

 

Using Hann the 3rd harmonic level is clearly wrong. 

Comparison DR = 48.55dB

Comparison THD+N = -27.32dB

Comparison THD   = -26.06dB
    H1 (1000Hz) = -4.5dB
    H3 (3000Hz) = -105.85dB
    H5 (5000Hz) = -73.23dB
    H7 (7000Hz) = -97.13dB

 

Using Dirichlet no H3 is listed.  There is one there in the spectrum at about -27 db. 

Comparison DR = 31.52dB

Comparison THD+N = -27.32dB

Comparison THD   = -41.64dB
    H1 (1000Hz) = -9.61dB
    H5 (5000Hz) = -53.17dB
    H7 (7000Hz) =  -71dB

 

This is Nutall:

Comparison DR = 53.53dB

Comparison THD+N = -27.32dB

Comparison THD   = -26.69dB
    H1 (1000Hz) = -3.39dB
    H3 (3000Hz) = -39.42dB
    H5 (5000Hz) = -77.8dB

 

This is Kaiser:

Comparison DR = 51.07dB

Comparison THD+N = -27.32dB

Comparison THD   = -28.26dB
    H1 (1000Hz) = -2.41dB
    H3 (3000Hz) = -29.8dB
    H5 (5000Hz) = -43.5dB
    H7 (7000Hz) = -70.78dB
    H9 (9000Hz) = -88.49dB

 

Here is Flat top windowing. 

Comparison DR = 50.84dB

Comparison THD+N = -27.32dB

Comparison THD   = -26.73dB
    H1 (1000Hz) =   -2dB
    H3 (3000Hz) = -29.37dB
    H5 (5000Hz) = -51.41dB
    H7 (7000Hz) = -106.46dB

 

ON music I typically use blackman-harris 7 or Hann.  

 

 

 

Blackman Harris 7 is a safe choice. Kaiser seems to want more samples for proper analysis. It is one of the better side-lobe rejecting windows in DW, but it doesn't work (right) if the FFT size is equal to the number of samples in the file.

 

EDIT: To clarify, it's not a property of the Kaiser window. It's the property of how it was implemented in DeltaWave (and so must be my error somewhere!)

 

[pkane2001]

1 minute ago, Arpiben said:

 

Hi Paul,

 

But at least you got the frequencies right , not yet my case. AP's frequencies aren't equally spaced.😉 

 

The frequencies are just the exact middle frequency of the corresponding FFT bin for  FFT size of 65536.

[pkane2001]

10 hours ago, Arpiben said:

 

Yes from my point of view, but I am not THD expert. It seems that algorithm calculates the difference between Ref&Comp harmonic at first glance. But there is probably some power density calculation within a specified bandwidth which might be tricked...No idea. Paul will help us 😉

 

The value computed by DW is THD+N. The algorithm is pretty simple: find the frequency with the largest amplitude, then apply a notch filter to remove it. What remains is the harmonic distortion and noise. Of course, this will only work for a simple sine wave test signal, anything more complex will require more complex filters.

[TomCapraro]

Here, I like it.
Exact results are obtained.
The signal is a 1000hz tone but with a periodic FFT 64 function activated (it is generated with REW)
The FFT window is the Dirichlet.

Sine_1000_96k_Float_LR.wav reccc.wav

[modmix]

On 10/26/2019 at 12:14 AM, esldude said:

So how do you get that display to come up in Deltawave.  Must be a key or hidden menu I missed. 

would like to learn that, too.
Ulli

[pkane2001]

2 hours ago, modmix said:

would like to learn that, too.
Ulli

 

That THD legend overlay doesn’t exit yet  but can be easily added if you think it’ll help.

[modmix]

Might be usefull in some cases, yes.
Would like to have it.
TIA

Ulli

[rando]

@pkane2001 How many keyboard shortcuts are there for this program?  For us dumkoffs that use it for actual musical comparisons and probably a few interested in the direction it is currently headed.  

 

 

Since your payoff here is feedback.  When you consider taking this market I wonder if you will be able to wrap up all the progressive thinking being applied back into a simple format.  I really like the idea of being able to overlay/hide items that don't require (or possible do) graphing by some unintrusive means. 

 

Very politely I would like to add I feel the addition of additional charts requiring arrowing was a small step backwards in design.  Perhaps assignable charts is an option that would for instance allow hiding spectogram on bit perfect files to make room for a specific chart or three on the far end.  You've done an admirable job of fighting clutter in the options panel and in the main program up until scrolling was introduced, IMO.  Just a thought I hope doesn't cause you (much) grief.

 

[esldude]

5 hours ago, modmix said:

would like to learn that, too.
Ulli

Currently that info is in the bottom of the results page if you have simple waveforms checked in the settings.

[pkane2001]

4 hours ago, rando said:

@pkane2001 How many keyboard shortcuts are there for this program?  For us dumkoffs that use it for actual musical comparisons and probably a few interested in the direction it is currently headed.  

 

 

Since your payoff here is feedback.  When you consider taking this market I wonder if you will be able to wrap up all the progressive thinking being applied back into a simple format.  I really like the idea of being able to overlay/hide items that don't require (or possible do) graphing by some unintrusive means. 

 

Very politely I would like to add I feel the addition of additional charts requiring arrowing was a small step backwards in design.  Perhaps assignable charts is an option that would for instance allow hiding spectogram on bit perfect files to make room for a specific chart or three on the far end.  You've done an admirable job of fighting clutter in the options panel and in the main program up until scrolling was introduced, IMO.  Just a thought I hope doesn't cause you (much) grief.

 

 

Just to make sure that you are aware (and I know it's not obvious) but you can already show/hide the charts as needed. You can do this directly from the settings screen or from the View->Tabs menu. Any charts that are hidden will not clutter the screen, and what's more, will not be computed saving processing time and making everything quicker.  Does this help?

 

or

[Arpiben]

On 10/26/2019 at 11:56 PM, pkane2001 said:

 

The frequencies are just the exact middle frequency of the corresponding FFT bin for  FFT size of 65536.

 

Here are the frequencies used by AP's 32 tones test (192 kHz sampling/FFT 65536).

IMHO, the distribution is not so obvious .😉

Rgds.

 

 

[TomCapraro]

I keep making tests.
Second test.👍

[TomCapraro]

Very good indeed.
I understand that below -110dB does not detect more harmonics, after all at those levels you can not even define distortion.
I also use gain alignment so that the comparison positions it at 0dBFS.

[pkane2001]

1 hour ago, Arpiben said:

 

Here are the frequencies used by AP's 32 tones test (192 kHz sampling/FFT 65536).

IMHO, the distribution is not so obvious .😉

Rgds.

 

 

 

Seems like some fraction of a decade increment (actually, looks like 1/10 of a decade):

 

[esldude]

7 hours ago, rando said:

@pkane2001 How many keyboard shortcuts are there for this program?  For us dumkoffs that use it for actual musical comparisons and probably a few interested in the direction it is currently headed.  

 

 

Since your payoff here is feedback.  When you consider taking this market I wonder if you will be able to wrap up all the progressive thinking being applied back into a simple format.  I really like the idea of being able to overlay/hide items that don't require (or possible do) graphing by some unintrusive means. 

 

Very politely I would like to add I feel the addition of additional charts requiring arrowing was a small step backwards in design.  Perhaps assignable charts is an option that would for instance allow hiding spectogram on bit perfect files to make room for a specific chart or three on the far end.  You've done an admirable job of fighting clutter in the options panel and in the main program up until scrolling was introduced, IMO.  Just a thought I hope doesn't cause you (much) grief.

 

Unless I am forgetting some there are only two hidden features. 

 

One is once you are looking at a spectrum, you can hold the shift key and click.  Then move the pointer to a portion of the spectrum which will show frequency and DB level.  Releasing leaves it there and you can do additional such pointers.  In post 1047 (a couple up thread) you can see where Tom has done this for several points. 

 

The other is holding down the scroll wheel and clicking lets you select an area to zoom in on a graph.  You can do that for a certain part of the graph and it will scale automatically.  Or you can do it along the edge of either axis to only change the one axis.  

 

Also if you didn't know, you can simply place the mouse in a graph and roll the scroll wheel to zoom in or out or place the mouse pointer along either edge axis and scroll the zoom in or out for that axis.  

 

Oh and you can put the mouse pointer in a graph and click and hold.  This lets you grab the graph and move around within it if you wish.  Also the same along either axis.  

 

If I forgot any, maybe others will add to my list. 

 

 

[esldude]

2 hours ago, Arpiben said:

 

Here are the frequencies used by AP's 32 tones test (192 kHz sampling/FFT 65536).

IMHO, the distribution is not so obvious .😉

Rgds.

 

 

I think I posted a link to a post on ASR in the last few days.  As well as those frequencies, they are set to different phases I assume to reduce peak levels.