24/352.8 vs. 16/44.1 - an attempt at a comparison

[manisandher]

I’m not really sure how much utility this thread is likely to have, but there are some things I’d like to pass by people. So here goes…

 

I’m going to use the following nomenclature to try to keep things clear:

file = original source material

- downloaded directly from 2L site

- not manipulated in any way, shape or form before replay

- very little music content above 10kHz

- one file is the original 24/352.8

- the other file is a 16/44.1, derived by 2L through decimation of the original 24/352.8

capture = digital file created by me, by capturing the analogue output of my DAC with an ADC

- ADC set to 16/44.1

- I used a Prism AD124 ADC, which was considered excellent in its day, but is perhaps not SOTA today

- the digital output from the Prism was fed via spdif to a Tascam DA-3000, which was used purely as a digital interface to a memory card

- one capture is the original 24/352.8 file being played back totally natively (i.e. no upsampling, filtering or SDM) in the software player

- the other capture is the 16/44.1 file being played back with 8x upsampling in the software player

 

In the “Some commonsense” thread, I posted a couple of 16/44.1 captures. Listening to the two original files directly, there was an obvious difference in sound, with the 24/352.8 being vastly superior to the upsampled 16/44.1. However, the two 16/44.1 captures sounded far more similar – though they still sounded audibly different to my ears.

 

@mansr, @pkane2001 and @Arpiben analysed the two captures, and pointed out that the difference in sound might be down to the particular upsampling filter I used when replaying the 16/44.1 file. (I really like the way this filter sounds, but it doesn’t measure very well in the frequency domain. Some would refer to it as a ‘leaky filter’.) So I decided to redo the captures, using a filter with better frequency-domain performance (the ‘poly-sinc' filter in HQPlayer, with TPDF dither). Here are the new captures:

 

https://drive.google.com/open?id=1u0CErRn-2XJQCnNZLdr7WILjgx4xJ2w1

 

https://drive.google.com/open?id=1d7C1LL_IuMdUyE5LP_3GjV9nl-U4W_NI

 

I think these captures sound very subtly different from each other (though I'm not sure I'd be confident doing a blind ABX). In any event, I’d like to understand a few things:

 

Could there be anything in the original 24/352.8 file that is lost during decimation down to 16/44.1?

If so, what could have been lost, considering there’s virtually no music signal above 10kHz anyhow?

Is all this hires malarkey really much ado about nothing?

 

I'm not sure if these two captures are going to be useful in helping to answer these questions, but feel free to take a listen, and/or perform any analysis.

 

Is it possible to identify which is the capture of the 24/352.8, and which is the capture of the upsampled 16/44.1?

 

Mani.

[manisandher]

13 hours ago, mansr said:

Here's a quick analysis.

 

Thanks Mans - that's excellent.

 

13 hours ago, mansr said:

The only anomalous thing here is the discrepancy between the DXD original and the 16/44.1 version. A plain downsample shouldn't differ that much. I wonder if perhaps they intentionally boosted the highs by a few dB for some reason.

 

How strange. Why on earth would they do this deliberately? I'm tempted to do my own decimation from 24/352.8 to 16/44.1 to see if I could do a more accurate job.

 

13 hours ago, mansr said:

As for the captures, the ADC has possibly been a limiting factor here. I would have used a 24-bit ADC at a higher sample rate to make certain nothing was missed.

 

I did think about this. I could have used the Tascam as the ADC, as opposed to simply using is as a digital interface. However, I really don't like the way it sounds, at any resolution - it measures well, but sounds cr*p. (Anyone who thinks most modern ADCs are essentially perfect obviously don't listen to them.) I do have another decent ADC here that is 24/192-capable. Will give it a go sometime.

 

Mani.

[manisandher]

5 hours ago, fas42 said:

There's a peculiar glitch in the recording chain, which also occurred in the first two samples manisandher put up, in the other thread,

 

That's no doubt down to the Tascam's auto-start function - it seems to get itself into a bit of a muddle. But this only lasts for a very short period of time, after which it sorts itself out. I'd discard the first second or so of all the captures, to be on the safe side.

 

Mani.

[manisandher]

11 hours ago, fas42 said:

Here's another interesting grab of the two waveforms, zoomed out on the time scale, with filtering to show only the bandwidth 8k to 16k,

 

 

T004 is blue, T005 is pink - does this make it easier to guess?

 

So what do these suggest Frank? That T004 is of higher resolution than T005? That T004 is more noisy than T005? Or perhaps that dithering is having an affect on the replay of the 16/44.1 file?

 

Mani.

[manisandher]

11 hours ago, fas42 said:

I certainly have a preference, though, .

 

What's your preference?

 

Mani.

[manisandher]

11 hours ago, sandyk said:

That's a pathetic excuse for not having gear (or hearing abilities ?) good enough to easily hear which file was the up-converted 16/44.1 version.

 

Alex, neither of the original files I used were 'up-converted'. One was the original 24/352.8 and the other a 16/44.1, derived by decimating ('down-converting') the original (done by 2L, not me).

 

One of the reasons I went for capturing at 16/44.1 was that it would be difficult distinguishing between the files by analysis alone... or so I thought.

 

11 hours ago, sandyk said:

You appear to come from the 16/44.1 camp that insists that high res is a waste of time as only younger dogs, bats etc. can hear the differences.

 

Both the captures I linked to were 16/44.1, so if they sound different, and they do (very subtly) to my ears, then this cannot be due to one having more HF info above 22.05kHz than the other. What it might be down to is the anti-imaging filter used when replaying the 16/44.1 file, as no anti-imaging filter was used in the case of the 24/352.8. (The anti-aliasing filter used by 2L during decimation may also be having an effect.)

 

Mani.

[manisandher]

17 hours ago, PeterSt said:

To prove this, Mani should (if he feels like it) try to take these two takes of exactly the same. It could happen that, now knowing this, he is able to make the "sensitivity" of the start of the recording such that there's hardly a time difference and that this will result in only a couple of difference like the two plots show, throughout, but I suppose that the difference in noise (which is random) at some degree of sensitivity (of recording start) will mangle with this.

 

Hey Peter, thanks for the suggestion. I'll take a look at how low I can take the auto-start sensitivity and then take a few dry runs.

 

Mani.

[manisandher]

1 hour ago, esldude said:

I know this is asking a lot Mani.  Could you repeat both captures at least once or twice?  Would be good to know how consistent the captures are.  And you could listen and see if the same capture twice or thrice all sound the same at least to each other.  And offering those re-captures for us to download too.  

 

Will do.

 

Mani.

[manisandher]

28 minutes ago, mansr said:

Hey goldenears, how does this one sound compared to the original DXD and 2L's down-conversion?

2l-092-1644.flac 3.44 MB · 1 download

 

'Platinum Ears' here 😀. Will take a listen tomorrow...

 

Mani.

[manisandher]

1 hour ago, pkane2001 said:

 

Mani, not sure I can add much more to the analysis that's already been posted. There are a few dB differences in frequency response above 8kHz or so, and what appears to be a very mellow filter applied at about 20k in T005 (in white):

 

Otherwise, the matching is pretty good with DeltaWave: rms null of about -81dB and correlated null of 84dB.

 

Phase error is fairly benign up to about 10k, but also increases with frequency:

 

 

I just added a new feature to DeltaWave to allow for listening to the ultrasonic differences  Decided to test it on these two files. Remember that the amplitude above 8k is low, and especially in a delta file the level is extremely low. I had to increase gain by +50dB in DeltaWave to hear the delta, and +40dB to hear the original files. You may need to do this in Audacity or another software.

 

https://drive.google.com/file/d/1JL_SAzw0-2mCMBqw7wj4FD1W83N4Ey_k/view?usp=sharing

 

See if you can spot any patterns/musical notes in it. Files are:

mani-T004-8k.wav - File T004 downshifted 8kHz

mani-T005-8k.wav - File T005 phase/amplitude aligned with T004 above, and also downshifted by 8kHz

mani-delta-8k.wav - difference between the two files above

 

In the delta file, I can only hear something above the noise at around 1:35.5 seconds, exactly where you can see the two spikes, below. There's a lot more audible in the T004-8k and T005-8k files, so there's definitely musical content above 8kHz

 

 

Here's the spectrum of T004 and T005 downshifted by 8kHz:

 

That's excellent Paul. Thanks for doing this.

 

I'll take a listen to the files later tonight, once I've put the kids to bed.

 

Mani.

[manisandher]

On 6/16/2019 at 9:03 PM, mansr said:

Hey goldenears, how does this one sound compared to the original DXD and 2L's down-conversion?

2l-092-1644.flac 3.44 MB · 15 downloads

 

Just compared your version to 2L's DXD and 16/44.1.

 

The DXD, played back with no upsampling, filtering or SDM, remains the clear winner for me. It simply sounds the nicest and the easiest to listen to.

 

Against 2L's 16/44.1, yours sounds softer, slightly fuller, almost as though the transients have been flattened a little. Not sure which I'd take given a choice. But they do sound audibly different to me.

 

Mani.

[manisandher]

On 6/17/2019 at 4:41 AM, Rexp said:

Ah so the point was to see if folks could hear a difference between the original hires and the downconverted 16/44. I guessed T004 was the 16/44, is that right? 

 

I could (and still can) hear a clear difference between the original DXD file (played back with no further upsampling, filtering or SDM, which 99.9% of DACs on the market will perform, to varying degrees), and 2L's down-converted 16/44.1 file (played back with 8x upsampling and filtering, but still no SDM).

 

If there were any point to this thread, it might be to figure how the two could sound different, considering there's very little signal above 10kHz. I decided to share a couple of 16/44.1 captures to help.

 

3 minutes ago, Rexp said:

So the take away is your ADC wasn't capable of capturing what was on the hirez file? 

 

Set to 16/44.1, as it was, it definitely wasn't capable of capturing everything that was on the hirez file. And yet the two captures still sound subtly different to me, but to a much lesser extent than the 2L files played back directly.

 

Mani.

[manisandher]

1 hour ago, mansr said:

Thanks for checking. I used shaped dither, lowering the noise by 10 dB below 15 kHz. Like this:

 

Did you apply this to the DXD during decimation, or directly onto the 16/44.1?

 

Mani.

[manisandher]

Your comparisons of the two captures look incredibly similar:

On 6/15/2019 at 10:11 PM, mansr said:

 

 

 

21 hours ago, pkane2001 said:

 

 

 

I know @pkane2001 is using his newly developed DeltaWave software, but what about you @mansr?

 

Mani.

[manisandher]

2 minutes ago, mansr said:

Do you suppose it could be the bit depth difference that's relevant here? Have you tried a 24/44.1 downconversion?

 

I suppose it's the most likely culprit. No, I haven't tried yet... but do intend to, when I have a bit more time.

 

Mani.

[manisandher]

21 hours ago, manisandher said:

If there were any point to this thread, it might be to figure how the two could sound different, considering there's very little signal above 10kHz. I decided to share a couple of 16/44.1 captures to help.

 

9 hours ago, fas42 said:

That's not the way to think about it. The signal above 10kHz is absolutely crucial...

 

Where did I imply or say that the signal from 10-15 kHz wasn't important? All I said was that there was very little of it above 10 Khz, as all the analyses have shown.

 

To my ears, the two captures sound very subtly different. Here's Paul's analysis once again:

 

On 6/17/2019 at 5:08 PM, pkane2001 said:

 

 

Do you think that the very subtle difference in sound that I'm hearing might be down to the difference between their respective FRs >10 kHz?

 

Mani.