Is a modified Offset audible in a .wav track?

[Empirical Audio]

Offset is the leading and trailing nulls in the .wav format before and after the music data. It is not part of the music and it does not create any "dead" time in the music track. It is simply part of the format. Given this, it would follow that this should not impact sound quality in any way, unless there is something at work here we don't yet fully understand.

The following track downloads contain 4 tracks of piano, 2 being 16/44.1 and two being 24/96. Each of the pairs contains one unaltered track and one track with 200 nulls added to the offset.

https://www.dropbox.com/sh/g9jz9lwgvozepic/AADx1d8YLCr5YCPQl23fUkFDa?dl=0

Since this is not an A/B/X test, one can only rate each of the tracks for things like dryness, ringyness, attack, decay, warmth, depth or shallowness of soundstage, clarity, wooliness etc. If you listen to these tracks, please rate them 1-4 with 1 being the most live sounding and 4 the least. Each track should have a unique number assigned, for example:

1) track 2a
2) track 1
3) track 2
4) track 1a

Also, please list your playback software, DAC and any re-clocking/resampling that occurs in your system. This will help answer the question of whether this is a hardware or software anomaly. Whether you use S/PDIF, network or USB to connect to the DAC.

Everyone does not have to get the same result to make this a valuable exercise. Even if they don't, if there is a definite trend that demonstrates that the tracks sound different, this is useful too.

After about 2 weeks, I will post all of the results. We will see if there is any smoking gun here. This test is already ongoing on several other forums, so I will combine all of those results as well.

Thanks,
Steve N.

 

[PierreV]

Just in case, you understand how the RIFF format is structured?

 

[BYRTT]

Vote:
FRAME1=2
FRAME2=3
FRAME1a=2
FRAME2a=4

FRAME2/FRAME1a sounds fine, but its like recording microphone is more close/hot which is not as natural as a normal listening seat several meters away and therefor lacks that natural room ambience, FRAME1/FRAME2a have this better difuse ambience that sounds as recording microphone distance is more natural to a listening seat some meters away which in my ears give a better natural sound, more difuse pleasing and tiny bit less resolution on notes than FRAME2/FRAME1a.

Playback software v23 64bit JRiver / R128 algorithm volume leveling / Equal loudness ISO226-2003 calibrated / track material rates are played native over WASAPI 32bit to Khadas Tone Board DAC. Tranducer domain for test was using head phones well a Neaurochrome HP-1 amp and DSP corrected HD650 cans.

Looks forward results and reveal down the road : )

 

[March Audio]

I thought we had already established/demonstrated/measured in the other thread that the audio data is identical.

If so the only "smoking gun" is that people are inconsistent judges of audio quality.

 

[KSTR]

The data is identical but the actual playback process may be different, with regard to the way the stream is packeted into SPDIF or USB packets, and that may result in a different jitter signature if the DAC used isn't up to the task. If the offset happens to be a multiple of the packet size, then the packets are again identical, as a special case. 200 samples is not a packet size (which most often is 2^N or k*2^N, k being a small integer).

That said, I cannot detect any audible difference induced by the offset (most likely also influenced by my personal confirmation bias which makes me believe any decent DAC, especially when USB-connected, can handle this with ease, plus any potential jitter signature would be arbitrary anyway).
Foobar2000/Audition3.0, no resampling --> ASIO (1024 samples buffer size) --> USB --> RME ADI-2 Pro FS --> Sennheiser HD700 (with some corrective EQ in the RME).

 

[BYRTT]

I thought we had already established/demonstrated/measured in the other thread that the audio data is identical.

If so the only "smoking gun" is that people are inconsistent judges of audio quality.

Ha ha that is good i'm one of the inconsistent then, that said i just did extract spectrum using Audacity for first 109,2 seconds and exported to REW to get them overlaid to look for differences and actual there is small variances here and there especially at tail ends.

44,1kHz tracks:

96kHz tracks:

 

[KSTR]

^ This is simple user error because you did not remove the heading 200 samples, and that leads to the analyzed sections not being the same, one section is missing 200 samples of actual audio data at the end. The spectra *must* be different just because of this. If you had choosen *corresponding* sections for analysis (say, samples 0 to 100000, and 200 to 100200) there is no difference as already confirmed by the simple subtraction test.

 

[BYRTT]

Thanks help analyze user error @KSTR will then try that and do a relistening tommorow : )

 

[KSTR]

Looks like Audicitity's FFT isn't the best, and/or something has gone wrong during the export to REW.

Checking specta with Audition produces identical graphs (96kHz variants) even *without* the offest corrected for analysis:

At least the differences are below one pixel's resolution.

 

[RayDunzl]

Looks like Audicitity's FFT isn't the best, and/or something has gone wrong during the export to REW.

Try harder...

When I align Frame1 and Frame2, and invert Frame2, then mix those tracks to a new track, I get a null result.

No difference in the data.

With a one-sample alignment error, I can get something that looks like yours:

 

[PierreV]

Guys, this is edging towards insanity at an accelerated pace...

Please check the screenshot I posted above...

Oh, and if you want to do listening tests, what about a file that has been modified by myself in Prayer Pro and now contains not only part of a stereophile article but also the start of Genesis (from the bible, not the band)

https://www.dropbox.com/s/0aaqfm7m83b4pma/frame1b.wav?dl=0

I should sound more stereophilic and angelic...

 

[March Audio]

aaawww.....no Genesis....but I am a closet prog rock fan

Not something you normally admit in public.

Thats it Im putting on Apocalypse in 9/8

 

[solderdude]

Can someone playback the file and record the audio output (high sample rate) and use Paul's software to compare the results.

I think it has been established the files are the same and should sound the same.
If there are any audible differences (which I have a hard time believing) it should be measured at the analog output of a DAC not in the file itself.

The same can be done to put the USB/SPDIF cable sounds different and has less bass or treble nonsense which by definition isn't possible.

 

[KSTR]

Please check the screenshot I posted above...

Sure, it tells us that the sample streams in both files have the same length. That is sort of a null info, though (correct but irrelevant).
It doesn't tell us that one of the streams is rotated (not: shifted) by 200 samples.

 

[KSTR]

Can someone playback the file and record the audio output (high sample rate) and use Paul's software to compare the results.

I might get to try that later, recording a lowly TASCAM US144/122 with the RME Adi-2 Pro as the recorder. I strongly prefer synced clocks which makes the diff test so much more robust and, more importantly, allows for block-averaging, but I'll probably try DW with async recording as well.

 

[PierreV]

Sure, it tells us that the sample streams in both files have the same length. That is sort of a null info, though (correct but irrelevant). It doesn't tell us that one of the streams is rotated (not: shifted) by 200 samples.

That info, I provided (along with others) in the other thread... but yeah you are right, I should have added a recap of the whole thing for clarity...
I guess I'll try to stay out of this before someone comes up with silver bits...

 

[Pluto]

Can someone playback the file and record the audio output (high sample rate) and use Paul's software to compare the results

OK – have done exactly that and confirm that Pkane's software reveals that, after correction of the deliberate offset, difference artefacts between the files are at least 50dB down (probably rather more in reality) from the main signal.

Nice piano playing although the mic. is rather too close, causing what seems like a decent, well-tuned, joanna to sound a bit clunky with mechanical noise in evidence.

If my understanding of the actual mission is correct. the goal is NOT to repair the offset and see if the two files sound the same, but to see if the offset itself causes a sonic difference owing to some aspect of the consequent packet handling. If the latter understanding is correct, I fail to see why the two file-pairs are at different sample and bit rates. If you wanted to determine if offset alone can make a difference, why not add padding to the original but otherwise retain the same data?

As it is, my answer is no audible difference and measured conversion artefacts much as I'd expect but of no audible significance.

 

[KSTR]

If my understanding of the actual mission is correct. the goal is NOT to repair the offset and see if the two files sound the same, but to see if the offset itself causes a sonic difference owing to some aspect of the consequent packet handling. If the latter understanding is correct, I fail to see why the two file-pairs are at different sample and bit rates. If you wanted to determine if offset alone can make a difference, why not add padding to the original but otherwise retain the same data?

Well, that's exactly what Steve did. "frame1a" is *identical* to "frame1b" in terms of the sample stream, only that one is rotated by 200 samples vs the other. Same with "frame1" vs "frame2", this time being 44.1/16 files.
Comparing the files yields nothing (by definition), we must look at the difference in the recorded analog output.

 

[Pluto]

Strikes me as a fairly futile experiment, I'm afraid. Anyway, no audible difference as I said.

we must look at the difference in the recorded analog output

I suspect that this would be as much a test of the test equipment itself as it would of any anomalous behaviour entirely down to the 200 sample offset!

I'm afraid that this experiment is far too close to the "life's too short" category to be of great interest (to me anyway).

Worst case, what might this demonstrate?

 

[KSTR]

... which i just did.

Alas, it looks like my old US-144mkII is broken (gotta see if I find the other one that I have somewhere). I get some strange random glitches, both when using local analog loopback as well as recording the output with the RME synced via SPDIF-out of the TASCAM, and the RME should easily cope with any SPDIF jitter.

I was still able to do some diffing on undisturbed sections, using an A-A' baseline diff compared to a A-B diff. A is recording of "frame1a" playback and "B is "frame2a", A' is a second take of "frame1a" for the baseline.
A-A' still is dominated by the plain pretty undistorted audio sligthly above the analog noise level (to be expected from minor gain drifts in both devices) but there are some semi-constant birdies at 4kHz multiples (actually more like 3900Hz). Semi-constant means sometimes their level drops to another constant level within some sample blocks.
In A-B the audio residual is about 3dB higher (and the background noise very similar), still explainable by gain drift, but this time the birdies are stronger, more like 6dB, and the 4kHz main bird is especially stronger and solid as a rock. Very audible in the diff after boosting the residuals by 60dB.

I don't know what to make out of this, given the skechy setup... but the N*3.9kHz components could well be USB packet noise even though the spikes should be at 1kHz and multiples....