Blue or red pill?

[pkane2001]

22 hours ago, manisandher said:

 

That looks like a really neat piece of software you've created there.

 

Could you compare the digital files 21 and 22 at some point please? It'd be interesting to see if you get the same result as @esldude.

 

Mani.

 

Mani, here's the comparison between 21 and 22:

 

Frequency spectrum of the two files after alignment:

 

And the actual remainder waveform, after subtracting the two:

 

I don't see the same spikes that @esldude reported, but I do see a slight difference in amplitude, which puzzles me if these were both digital captures with no changes upstream except for the SFS setting.

 

[PeterSt]

1 hour ago, pkane2001 said:

which puzzles me if these were both digital captures with no changes upstream except for the SFS setting.

 

Paul, Or your software is not correct, or Mani did not apply the correct procedure, or something else is going on.

I tend to even vote for the latter. 

[pkane2001]

Just now, PeterSt said:

 

Paul, Or your software is not correct, or Mani did not apply the correct procedure

 

I would guess my software, except that I see this difference before doing any processing. It is a tiny difference, but consistent. I’m going to double-check the logic to make sure.

[manisandher]

2 hours ago, pkane2001 said:

Mani, here's the comparison between 21 and 22:

 

Thanks Paul.

 

2 hours ago, pkane2001 said:

... but I do see a slight difference in amplitude, which puzzles me if these were both digital captures with no changes upstream except for the SFS setting.

 

Yes, this is strange. I made no changes other than the bit-identical SFS setting in XXHighEnd.

 

But as I've mentioned before, I was slaving the Tascam to the MOTU in the hope that this would help with analysis. However, perhaps this is 'illegal' - the Tascam was receiving an spdif signal from the audio PC and perhaps required its own internal wordclock to reclock this recover the clock from the spdif stream. It's strange though because the Tascam didn't report a wordclock error (it's usually very quick to do this) and seemed perfectly happy with the arrangement.

 

Edit: The two digital captures (21 and 22) sound identical to me, and I can't hear anything obviously untoward in the way they sound.

 

Mani.

[pkane2001]

16 minutes ago, manisandher said:

 

Thanks Paul.

 

 

Yes, this is strange. I made no changes other than the bit-identical SFS setting in XXHighEnd.

 

But as I've mentioned before, I was slaving the Tascam to the MOTU in the hope that this would help with analysis. However, perhaps this is 'illegal' - the Tascam was receiving an spdif signal from the audio PC and perhaps required its own internal wordclock to reclock this recover the clock from the spdif stream. It's strange though because the Tascam didn't report a wordclock error (it's usually very quick to do this) and seemed perfectly happy with the arrangement.

 

Edit: The two digital captures (21 and 22) sound identical to me, and I can't hear anything obviously untoward in the way they sound.

 

Mani.

 

Mani, they look identical to me as well, I can’t imagine that they would sound different. The slight level difference is definitely below audibility, I’m just not sure where it comes from since theoretically the same exact samples were captured in both Wav files.

 

I’ll try to do a bit by bit comparison to see where there may be differences (or if this is just some rounding error somewhere).

[fas42]

What happened with the digital captures is that sync with the input was lost, being controlled by an alternative clock. Interestingly, the result was that, at least in one portion of the waveform, that the two digital captures were perfectly in step with each other, then lost the lock, only to re-establish it - with a regular beat. This looked like,

 

 

The flat line links are in fact perfect nulls, the two digital captures match in these intervals.

 

[STC]

On 4/30/2018 at 11:34 AM, Audiophile Neuroscience said:

 

Okay, color me intrigued.

 

 

Okay...I think I will reveal how I hear the difference.  Firstly, there is a slight tonal difference which IMO is due to the noise level. Secondly, the is a slight difference in the tempo. I cannot be consistent. In my earlier attempt where I got 9/10 wrong, I was going for tonal difference although I know there was a  slight difference in the soundstage but that was difficult to identify in ABX. In the second attempt I was concentrating on the tempo. I think Mani also mentioned about the first few piano notes being enough. Yes, I too relied the first few seconds. From 3 to 10 seconds window.

 

 

In one of my earlier posts, I requested Mani to record few samples with the two SFS so that I can show that changing of SFS can affect the tempo and length of the playback time. I am not sure if this can be proven but I did a lot experimentation with buffer size of the JRiver during my initial zoning experiment with the NUC. During those trials I did perceive that the tracks with larger buffer appeared slower to my ears. I never bothered to dwell into this any further as the difference is toooooo minor to matter and wasn't an issue anymore after the PC upgrade. Okay...now back to the bitter pill.

 

I do not have XXHE so I cannot confirm my observation with SFS settings. As an alternative I used Foobar buffer and recorded directly to Audacity of the Foobar output.

 

I made two recordings of  1) random X track 2) impulse response 3) Mani's digital A with 50ms buffer and 1000ms buffer.

I added a random and impulse response is create a marker since the manual start and stop made it impossible to time sync Digital A wave alone. 

 

Below are the Audacity screenshots of what can be seen from the analogue samples.

 

 

Notice the noise level during the silence between the two tracks.  In the 50ms buffer the noise level was higher than the 1000ms buffer.

 

_____________________________________

 

 

 

 

In this screenshot, the the Digital A started about 0.033s earlier with 50ms buffer setting. I have tried my best to time align but I leave it to the experts to correct me on this.

 

_______________________________________________________

 

 

Here I have time aligned digital A files by removing the track X and Impulse response so that I could measure the lenght of digital A itself with the two different buffer setting.

 

___________________________________________________________________

 

 

 

After time aligning the beginning of the track of Digital A, you can see that with buffer of 1000ms it took and extra 0.038s to complete.  This is why maybe I feel the tempo is slightly different with bigger buffer as I mentioned earlier.

 

_______________________________________________________________________________

 

 

 

Showing the whole length of Digital A after time aligning the start of Digital A.

 

_____________________________________________________________

 

 

All three tracks capture here before discarding the random X and realign from the impulse response.

 

___________________________________________

 

Ok guys go easy on me. I do not know much about this things. Just trying to show something what I observed so far.

 

 

 

 

 

 

 

 

 

 

[Audiophile Neuroscience]

54 minutes ago, STC said:

 

 

Okay...I think I will reveal how I hear the difference.  Firstly, there is a slight tonal difference which IMO is due to the noise level. Secondly, the is a slight difference in the tempo. I cannot be consistent. In my earlier attempt where I got 9/10 wrong, I was going for tonal difference although I know there was a  slight difference in the soundstage but that was difficult to identify in ABX. In the second attempt I was concentrating on the tempo. I think Mani also mentioned about the first few piano notes being enough. Yes, I too relied the first few seconds. From 3 to 10 seconds window.

 

 

In one of my earlier posts, I requested Mani to record few samples with the two SFS so that I can show that changing of SFS can affect the tempo and length of the playback time. I am not sure if this can be proven but I did a lot experimentation with buffer size of the JRiver during my initial zoning experiment with the NUC. During those trials I did perceive that the tracks with larger buffer appeared slower to my ears. I never bothered to dwell into this any further as the difference is toooooo minor to matter and wasn't an issue anymore after the PC upgrade. Okay...now back to the bitter pill.

 

I do not have XXHE so I cannot confirm my observation with SFS settings. As an alternative I used Foobar buffer and recorded directly to Audacity of the Foobar output.

 

I made two recordings of  1) random X track 2) impulse response 3) Mani's digital A with 50ms buffer and 1000ms buffer.

I added a random and impulse response is create a marker since the manual start and stop made it impossible to time sync Digital A wave alone. 

 

Below are the Audacity screenshots of what can be seen from the analogue samples.

 

 

Notice the noise level during the silence between the two tracks.  In the 50ms buffer the noise level was higher than the 1000ms buffer.

 

_____________________________________

 

 

 

 

In this screenshot, the the Digital A started about 0.033s earlier with 50ms buffer setting. I have tried my best to time align but I leave it to the experts to correct me on this.

 

_______________________________________________________

 

 

Here I have time aligned digital A files by removing the track X and Impulse response so that I could measure the lenght of digital A itself with the two different buffer setting.

 

___________________________________________________________________

 

 

 

After time aligning the beginning of the track of Digital A, you can see that with buffer of 1000ms it took and extra 0.038s to complete.  This is why maybe I feel the tempo is slightly different with bigger buffer as I mentioned earlier.

 

_______________________________________________________________________________

 

 

 

Showing the whole length of Digital A after time aligning the start of Digital A.

 

_____________________________________________________________

 

 

All three tracks capture here before discarding the random X and realign from the impulse response.

 

___________________________________________

 

Ok guys go easy on me. I do not know much about this things. Just trying to show something what I observed so far.

 

 

 

 

 

 

 

 

 

 

 

Hi ST

I'll let the technical experts deliberate on your cause and effect scenario/s but just to say nice detective work. Also, no one can dispute the "effect" given your ABX results.

[PeterSt]

2 hours ago, STC said:

Secondly, the is a slight difference in the tempo.

 

This is from only a few weeks ago the observation of an XXHighEnd user :

 

Quote In order not to repeat the same adjectives that others, I will say that once again I keep with the new settings. Just point out that I agree that there seems to be a better dynamic and sometimes I have the strange feeling like that the sound is going faster. (Re: 2.10 sound quality)

 

This remark is indirectly about a faster response with a certain settings combination (this is very easy to hear). So the faster (transient) responding settings imply a faster tempo to the listener. Of course this is not true for real, but logic once we see that faster transient gives a faster experience. I myself - I think that this can be due to micro transient matters which transients now pop up, while otherwise they are not observed at all. So say that a drummer lets dance his stick on a beat twice on the skin but the second hit arrives so quickly after the first that the first may be buried which a less transient responding system,  then unveiling that first hit literally makes the tempo faster because it just is (not talking about main beats of course).

PRaT is better.

 

 

 

 

[STC]

47 minutes ago, PeterSt said:

Of course this is not true for real, but logic once we see that faster transient gives a faster experience.

 

I am not sure what you are trying to say but from the analogue capture from the the beginning of impulse response and to end of Digital A is :-

 

1) 50ms buffer = 1733090 samples

2) 1000ms buffer = 1736450 samples.

 

Different is 3360 samples longer during playback with 1000ms buffer.  I am not sure whats the calculation should be but it should be like 3360 time 1/96000 which is about 35 milliseconds longer. 

[PeterSt]

1 minute ago, STC said:

I am not sure what you are trying to say

 

1 minute ago, STC said:

Different is 3360 samples longer during playback with 1000ms buffer.  I am not sure whats the calculation should be but it should be like 3360 time 1/96000 which is about 35 milliseconds longer.

 

In that case this is all moot (obviously). So what I understand now is that you say that the file is just played "stretched" with a different Foobar setting. Well, hooray for Foobar, but this is not really the idea. 

So what I understand only now is that you could not understand me because you odhered the idea that this kind of "stretching" is allowed. But it really is not. And btw, this is most certainly not happening in XXHighEnd.

Now you know this, you can try to reread my previous post and you will get the idea.

[STC]

8 minutes ago, PeterSt said:

 

 

In that case this is all moot (obviously). So what I understand now is that you say that the file is just played "stretched" with a different Foobar setting. Well, hooray for Foobar, but this is not really the idea. 

So what I understand only now is that you could not understand me because you odhered the idea that this kind of "stretching" is allowed. But it really is not. And btw, this is most certainly not happening in XXHighEnd.

Now you know this, you can try to reread my previous post and you will get the idea.

 

I am not stretching anything. The change in buffer slows the playback depending on the buffer size. I wanted to see if SFS is also doing the same.

[manisandher]

8 hours ago, fas42 said:

What happened with the digital captures is that sync with the input was lost, being controlled by an alternative clock. Interestingly, the result was that, at least in one portion of the waveform, that the two digital captures were perfectly in step with each other, then lost the lock, only to re-establish it - with a regular beat. This looked like,

 

 

The flat line links are in fact perfect nulls, the two digital captures match in these intervals.

 

 

Hopefully @pkane2001 will be able to verify this too with his new software. (It'll act as a good test that his software is working correctly.)

 

The reason why I slaved the Tascam to the MOTU was to help in aligning the analogue captures with their respective digital captures - so analogue 23 with digital 21, and analogue 24 with digital 22. I wonder if there's enough time during the nulls for this still to be possible? Might prove interesting...

 

Mani.

[manisandher]

3 hours ago, STC said:

I made two recordings of  1) random X track 2) impulse response 3) Mani's digital A with 50ms buffer and 1000ms buffer.

I added a random and impulse response is create a marker since the manual start and stop made it impossible to time sync Digital A wave alone. 

 

3 hours ago, STC said:

As an alternative I used Foobar buffer and recorded directly to Audacity of the Foobar output.

 

3 hours ago, STC said:

Below are the Audacity screenshots of what can be seen from the analogue samples.

 

I'm not sure I understand what you did.

 

The first two quotes suggest you were capturing digitally. The third suggests you were capturing the analogue output of your DAC.

 

Can you help me understand?

 

Mani.

[manisandher]

8 minutes ago, STC said:

I am not stretching anything. The change in buffer slows the playback depending on the buffer size. I wanted to see if SFS is also doing the same.

 

SFS does not slow-down/speed-up playback in any way, shape or form. If it did, the earlier digital captures (before I started messing around with slaving clocks) would not have nulled perfectly - but they did null perfectly.

 

I have no idea what Foobar is doing...

 

Mani.

[STC]

1 minute ago, manisandher said:

 

 

 

I'm not sure I understand what you did.

 

The first two quotes suggest you were capturing digitally. The third suggests you were capturing the analogue output of your DAC.

 

Can you help me understand?

 

Mani.

 

I was thinking about this too. I am capturing what’s coming out during the playback from the Foobar fed into Audacity. I think this is still in the digital domain. Even then there is a difference in the total samples that was recorded in the Audacity. 

[STC]

18 minutes ago, manisandher said:

 

SFS does not slow-down/speed-up playback in any way, shape or form. If it did, the earlier digital captures (before I started messing around with slaving clocks) would not have nulled perfectly - but they did null perfectly.

 

I have no idea what Foobar is doing...

 

Mani.

 

 

Your original  digital capture _ A  and  digital capture _ B is different by 22882 samples. How come they can be identical? 

[manisandher]

3 minutes ago, STC said:

Your original  digital capture _ A  and 1. digital capture _ B is different by 22882 samples. How come they can be identical? 

 

That's simple. Mans stopped playback at slightly different times during the ABX. Hence the digital captures were slightly different in length.

 

Once correctly aligned, they sit on top of each other perfectly. One just happens to go on for a slight bit longer after the other has stopped.

 

No stretching involved whatsoever.

 

Mani.

[STC]

Just now, manisandher said:

 

That's simple. Mans stopped playback at slightly different times during the ABX. Hence the digital captures were slightly different in length.

 

Once correctly aligned, they sit on top of each other perfectly. One just happens to go on for a slight bit longer after the other has stopped.

 

No stretching involved whatsoever.

 

Mani.

 

 

Oh ..yes. I forgot about one being longer. 

[STC]

1 hour ago, STC said:

I am not sure what you are trying to say but from the analogue capture from the the beginning of impulse response and to end of Digital A is :-

 

1) 50ms buffer = 1733090 samples

2) 1000ms buffer = 1736450 samples.

 

Different is 3360 samples longer during playback with 1000ms buffer.  I am not sure whats the calculation should be but it should be like 3360 time 1/96000 which is about 35 milliseconds longer. 

 

Just did another capture by increasing the buffer to the maximum of 30000ms. And the difference is samples between 50ms buffer and 30000ms is 7007 samples longer in 30000ms buffer.  Now let me copy this directly and see if there can be difference in the total samples.

 

edit: the digital copy is exactly 2672543 samples. Both start and end precisely at the same time. But looks like the samples captured after foobar lost quite a big number of samples. 

 

And the difference between 50ms buffer and original file Digital A is 116931 samples. I am not sure how to the calculation as the original files were in 176kHz sampling rate and the captures were in 96kHz.

 

[mansr]

13 hours ago, fas42 said:

What happened with the digital captures is that sync with the input was lost, being controlled by an alternative clock. Interestingly, the result was that, at least in one portion of the waveform, that the two digital captures were perfectly in step with each other, then lost the lock, only to re-establish it - with a regular beat. This looked like,

 

 

The flat line links are in fact perfect nulls, the two digital captures match in these intervals.

That's more or less what I'd expect to get with unsynchronised clocks.

[pkane2001]

5 hours ago, manisandher said:

 

Hopefully @pkane2001 will be able to verify this too with his new software. (It'll act as a good test that his software is working correctly.)

 

The reason why I slaved the Tascam to the MOTU was to help in aligning the analogue captures with their respective digital captures - so analogue 23 with digital 21, and analogue 24 with digital 22. I wonder if there's enough time during the nulls for this still to be possible? Might prove interesting...

 

Mani.

 

I can confirm the occasional clock error. It's a glitch, rather than a constant drift, so the clocks re-sync after a short period and then un-sync again. Here's what the difference of 21 and 22 looks like, with no processing:

 

 

And here's the same difference after DeltaWave did an alignment (no drift correction as no drift was found):

 

By the way, the misplaced samples is what I think caused the slight difference in gain shown in the FFT plot. I'll have to see if there's a better way to compute it, as I was basing it on a total energy computation. 

[mansr]

30 minutes ago, pkane2001 said:

I can confirm the occasional clock error. It's a glitch, rather than a constant drift, so the clocks re-sync after a short period and then un-sync again.

That's expected. The receiver will work until the phase difference becomes too great and it slips, re-locks, and starts the cycle anew. Any further analysis of those captures is meaningless as they do not accurately reflect what the DAC received.

[manisandher]

1 hour ago, pkane2001 said:

I'll have to see if there's a better way to compute it, as I was basing it on a total energy computation. 

 

1 hour ago, mansr said:

Any further analysis of those captures is meaningless as they do not accurately reflect what the DAC received.

 

Agreed.

 

From the outset, the digital captures have only ever been taken to ensure we're getting bit-identical playback. Based on all the previous digital captures, I think we can assume that the analogue captures 23 and 24 did indeed receive bit-identical data, even though the digital captures 21 and 22, taken at the same time, are full of errors. If 23 and 24 bring up something interesting, and we really need to verify bit-identical input, I'll set things up again and redo them.

 

Mani.

[pkane2001]

52 minutes ago, manisandher said:

 

 

Agreed.

 

From the outset, the digital captures have only ever been taken to ensure we're getting bit-identical playback. Based on all the previous digital captures, I think we can assume that the analogue captures 23 and 24 did indeed receive bit-identical data, even though the digital captures 21 and 22, taken at the same time, are full of errors. If 23 and 24 bring up something interesting, and we really need to verify bit-identical input, I'll set things up again and redo them.

 

Mani.

 

Mani, 

 23 and 24 seem to align well to each other. There's still a difference in frequencies above 17KHz, like I saw with other A/B analog captures earlier:

 

Here's the section where there are differences, zoomed in: