[KinGensai]

Ok... I have read as much as I can about this topic, and I think I have at least a cursory understanding of sampling rates & bit depth. The theory and practice makes sense to me, reproducing waveforms correctly up to 20kHz with a 12.5% operating buffer in 16/44.1 seems valid, and up until recently sound.

I dabbled in upsampling via UAPP and oversampling via Neutron for about half a decade at this point, and have never heard a difference up until now. The AES article on CD sound, Vnandor's post on 176k vs downsampled 44.1k, and the scattered discussions I have read here over time have and still make sense, but what doesn't make sense is the sudden change in my perception on upsampling.

For context, my current source is a V60 running in Aux mode powering my QDC V14 & Fatfreq Maestro SE. When trying upsampling, I turn off EQ and crossfeed, have the Direct and direct PCM flags checked, then flip between bit perfect playback and upsampled playback (384kHz).

https://www.hypethesonics.com/dapti-database?LG_V60

Before this setup, no audible change. Now, there is a very subtle change in complex sections of music. It's difficult to pinpoint without measurement equipment, but I perceive, for lack of a better term, smoother playback. The resonance peaks that usually cause my ears to tense up don't, and sounds that are buried in the mix are more noticably prominent. Honestly, it could just be within listening margin of error, and I can't tell if it's "better" or "worse", could it be compression?

Test tracks: 誰が為に and The Black Death Mansion Murders in 24/48 FLAC, acquired from ototoy.jp




I'm not sure what to think on this subject now. Sighted bias maybe? But my bias has been no audible difference for 5 years. Is it something UAPP is doing? Is the ES9219 chipset doing something different here? Maybe my equipment to this point just hasn't had sufficient technical performance for me to notice until now?

If you would like the source files without having to go to ototoy, let me know and I will send you the files I have.

 

[VNandor]

You could take a bunch of approaches here. If you have a way to capture the digital output of your device, you could record the regular and the upsampled output, use a high quality sample rate conversion software to downsample the upsampled version and then perform a null test. This strictly only tests two things: how accurately your phone upsampled the file and how accurately you downsampled it. This could miss some things, as an example, maybe your phone might play back higher sample rate files differently than lower sample rate ones and that would be something this test wouldn't catch.

You could capture the analog signal of your phone's output as well. Instead of plugging it into the amp, you plug it into a line in of your PC/audio interface to capture it. Then, you could do a null test for that as well. This would let you catch if your phone plays back different sample rate files differently but it brings their own set of problems that makes the null test hard to do manually. If you do it this way, I recommend using deltawave for the null test.

Finally, you could try doing a single blind test with the help of someone. Of course, switching will take a considerable amount of time but does a difference really matter all that much to you if it's forgotten in ~10 seconds?

I've just done the first test I've outlined with my own gear, I got a -54dB RMS null with a peak of -32dB. This doesn't sound all that impressive and certainly leaves some doubt about the conversion. However, I low passed this difference file at 19kHz, which knocked down the RMS to -95dB and the peak to -82dB. What this implies is that most of the differences are above 19kHz. The -54dB RMS and -32dB peak isn't a massive difference but it's not a definitve won't be ever heard number either. What makes it a won't ever be heard difference is the fact it's quite small to begin with and it comes from frequencies above 19kHz.

I don't think a sample rate conversion could ever apply a dynamic range compression to the signal. I could see it lowering the overall volume though, which could cause a similar feel to what you've described.

 

[KinGensai]

Ok, I tried recording two instances of about 15 seconds of the second track. Got them loaded up in Deltawave, had to align the upsampled recording cause I'm a humie or whatever. What should I do with this now?

https://mega.nz/folder/BoI1ULDZ#5RwjS4-XuQTsiujvJzSAtg

This has the report file Deltawave generated. I also included the recordings of the bit perfect playback and the upsampled playback.

Just from my layman's perspective, the squiggly lines look like they confirm what I hear... I think? I don't know. It certainly seems like everything got a bit quieter, not just the peaks.

 

[gregorio]

what doesn't make sense is the sudden change in my perception on upsampling.

There are two possibilities, either it’s just a change in your perception or there’s an actual audible difference that you’re hearing. Just upsampling should not make any audible difference, however, in the case you describe there are potentially numerous variables at play and an actual audible difference cannot be as easily dismissed as in most other scenarios. In addition to just the upsampling, you’re bypassing the default OS playback, which may include things like loudness normalisation or limiting. Even with just the upsampling, differences can occur between how the software and how your DAC chip deals with ISPs (inter-sample peaks) which could cause a small difference in output level. There are various other variables at play, because the V60 uses a sophisticated SoC and so various parts of it maybe be bypassed or brought into play if you’re feeding it an already oversampled signal (@ 32bit float?).

I’d second VNandor’s suggestion of a null test on the analogue output using deltawave.

Ah, I see you’ve done a null test with Deltawave, although I’m not quite sure what you’ve compared. Is that the analogue output or just the digital files before conversion but after the upsampling, I’m guessing the latter? It appears to show a difference in level of around 1.5dB, quite a difference in phase and in the higher freqs in the spectrogram. The latter two differences suggest a minimum phase filter with a slow roll-off may have been applied?

G

 

[KinGensai]

I connected my V60's analog out to my PC's mic in, then recorded the output in WAV. The file labeled 24\48 BP is bit perfect mode, the other is the UAPP upsampled version.

Phase differences may be user error, they are two separate recordings, so I matched the starting pulse as best I could, I'm not sure how to fix that perfectly.

 

[VNandor]

Ok, I tried recording two instances of about 15 seconds of the second track. Got them loaded up in Deltawave, had to align the upsampled recording cause I'm a humie or whatever. What should I do with this now?

https://mega.nz/folder/BoI1ULDZ#5RwjS4-XuQTsiujvJzSAtg

This has the report file Deltawave generated. I also included the recordings of the bit perfect playback and the upsampled playback.

Just from my layman's perspective, the squiggly lines look like they confirm what I hear... I think? I don't know. It certainly seems like everything got a bit quieter, not just the peaks.

Oh, I only used deltawave to get the difference file. Deltawave automatically aligns the files with subsample accuracy, aligns the gain, and compensates for clock drift. Doing these manually with simple/free audio editors is basically impossible as far as I know, except the gain matching. You can export the difference file by going to File ( in the top left corner)->Audio Data->Save delta. The auto gain match feature negates any volume difference that might be caused by the upsampling which could be the thing that you are actually hearing. Deltawave adjusted the gain by 0.52dB which might be just enough to be heard but not enough to register as a simple volume difference.

The difference (after the automatic processing) is -62dB RMS with a -39dB peak. After I lowpassed the difference file, the RMS dropped to -64dB, and the peak to -41dB. After normalizing the file I could listen to it and it sounds and looks like most of this difference is noise and high frequency content but I can't think of a way to actually quantify the split between them. To me personally, these numbers look low enough to not worry about it considering at least some of them comes from just noise.

Something else you could do is do a null test of the same modes, essentially you record them twice and compare those. This would reveal how resolving the test method is. Ideally you would get a perfect null in this case but of course that's impossible. If they don't null better than the different modes being compared then there aren't any differences that the test can show between the two modes apart from the ~0.5dB level difference.

 

[VNandor]

It appears to show a difference in level of around 1.5dB, quite a difference in phase and in the higher freqs in the spectrogram. The latter two differences suggest a minimum phase filter with a slow roll-off may have been applied?

I checked the provided A/B files directly and after trimming them to the same length (of the same section) their RMS differs about 0.5dB so I think deltawave applied the appropriate amount of gain matching before producing the difference file.

 

[KinGensai]

I ran the null test with two instances of bit perfect playback, and as far as I can see it's pretty much a perfect match.

https://mega.nz/folder/R5oxCIIK#T_6TcMihWKPChJuDTmH7Vw

Looks like I wasn't a crazy bat and was actually hearing something different, although it looks like it's just a slight level difference. False alarm I guess lol.

At least I'm familiar with what deltawave and audacity is now, I'm sure that'll be helpful should something like this come up again.

 

[KinGensai]

So, I have a subsequent question on this. I watched a video explaining what upsampling is in mathematical terms. So as I understand, upsampling and oversampling is the same process, it just denotes whether the DAC itself or the software is performing the interpolations.

My question is if these interpolation algorithims are standardized or if it's a "secret sauce" that sets hardware/software apart from each other. I'm still trying to figure out how 5k for a Chord M Scaler makes sense.

 

[gregorio]

I ran the null test with two instances of bit perfect playback, and as far as I can see it's pretty much a perfect match.

Actually there’s still quite a significant difference. The results strongly indicate a different anti-image filter. Although this difference would be inaudible as it’s pretty much all above 19kHz.

At least I'm familiar with what deltawave and audacity is now, I'm sure that'll be helpful should something like this come up again.

Indeed, the null test is a relatively ancient test but is still widely employed and useful for all sorts of things. Deltawave is helpful because it automatically avoids many of the potential pitfalls of null tests (such as time alignment) and also includes a function for “audibility”, by applying hearing determinants (such as auditory masking), that could otherwise result in a difference interpreted as well within audibility.

So as I understand, upsampling and oversampling is the same process, it just denotes whether the DAC itself or the software is performing the interpolations.

Not necessarily. Although both terms are commonly used interchangeably, they technically have somewhat different meanings. “Upsampling” indicates increasing the sample rate, while “oversampling” indicates increasing the sample rate to a multiple of the base rate. So increasing the sample rate of a 44.1kHz file to 48kHz is upsampling but not oversampling but increasing it to 88.2kHz is both upsampling and (2x) oversampling. The terms do not differentiate between whether a DAC or standalone software performs the task, although generally ADCs/DACs oversample rather than upsample. Also, the oversampling process in ADCs/DACs typically also involves a reduction in the number of bits (as bit depth and sample rate are interchangeable at high oversampling rates).

My question is if these interpolation algorithims are standardized or if it's a "secret sauce" that sets hardware/software apart from each other.

Neither. The up/over sampling algorithms are not standardised and are also not a “secret sauce“. Unless something is seriously broken/defective there will not be any audible difference between any of the various methods of over/up sampling.

I'm still trying to figure out how 5k for a Chord M Scaler makes sense.

It doesn’t make sense from an audibility point of view. It only makes sense from a “brand name”, visual appearance or “inaudible specs” point of view.

G

 

[KinGensai]

Is it common for an anti-image filter to change between sessions? I generally don't pay too much attention past 19kHz because that's getting really close to the nyquist frequency so is going to be subject to more variation from interpolation. I'm only concerned if I hear aliasing, and since I'm not engaging in the production side, I almost never do.

Neither. The up/over sampling algorithms are not standardised and are also not a “secret sauce“. Unless something is seriously broken/defective there will not be any audible difference between any of the various methods of over/up sampling.

It doesn’t make sense from an audibility point of view. It only makes sense from a “brand name”, visual appearance or “inaudible specs” point of view.

G

Yeah, I figure that's the case. Maybe it's doing something similar to what I experienced here? I'm not about to drop 5k to confirm that though lol.

 

[redrol]

I let my DAC do the heavy lifting here knowing that some engineer understands this better than my dumb self.

 

[Blackwoof]

Doing 48KHz with high bit rate MP3 seems really help with problem samples, It probably the only case where up-sampling is a benefit.

 

[redrol]

No it doesn't.. how could that do anything at all?!

 

[KinGensai]

Doing 48KHz with high bit rate MP3 seems really help with problem samples, It probably the only case where up-sampling is a benefit.

Could you elaborate?

Are you claiming that converting a 16/44.1 lossless file to a 24/48 MP3 helps in some way?

I don't think it does, the difference between 22.05kHz and 24kHz as frequency ceilings only really benefits in mitigating aliasing on digital equipment as well as infinitesimally small accuracy changes to the waveform approaching the nyquist frequency. It should not change anything in playback, and I have never heard anything but clipping from upsampled lossless -> lossy encodings.