Perhaps relevant enough to cause a noticeable difference in calculated playback gain when sent through a broken volume leveler?

Quote from: greynol on 2017-04-25 18:19:49

Perhaps relevant enough to cause a noticeable difference in calculated playback gain when sent through a broken volume leveler?

I don't know how Arnold made his test but I took Pink Floyd - Wish You Were Here DSD file and converted it to 352800 Hz PCM. This file showed RMS level at -18.39 dB. Filtering away frequencies below 22 kHz showed RMS level for the ultrasonics at -22.28 dB. And after resampling to 44.1 kHz to remove the inaudible sounds RMS level was at -20.68 dB.
At least with this this file and with this much ultrasonic content RMS is not a valid method for loudness matching. Computing ReplayGain for a 44.1 kHz or 48 kHz resampled version gives better results.

I've never understood the point of using old analogue recordings which are never going to use more than 13 bits or need a sampling rate greater than 48kHz.

Wasn't this used to try and discredit the Meyer-Moran SACD test?

g=938570 date=1493059974]

Quote from: Arnold B. Krueger on 2017-04-24 19:30:15

(2) Measure the RMS or if no RMS calculation is available, the average value associated with the level of music between the two points in each piece of music.

He should probably lowpass to no greater than the nyquist of the PCM samplerate first.

This is not necessarily as easy as you make out. My first post on HAudio was in the ABX section for Foobar, which was making comparisons impossible as ReplayGain wouldn't set the levels between PCM and DSD correctly (possibly due to the noise, but I won't speculate as I get told off for doing that).

So I HAVE tried a proper listening test, and the technology failed me.

Quote from: judd on 2017-04-25 07:59:50

FYI, my converter ReSampler is a command-line tool which can convert DSD .dsf and .dff files to various PCM formats. (Or if you prefer a graphical interface, use ferocious)

Maybe a bit off-topic but it seems that your software's UI is not DPI-aware. While hi-res audio does no harm to my ears a hi-res monitor does hurt my eyes. The recent Windows 10 Creators Update advocating improvement in UI scaling is a pure joke, I used it for no more than 3 hours and reverted to Windows 7.

Quote from: judd on 2017-04-25 07:59:50

files from the 2L Hires Test-Bench

Back to the topic. Just in case shakeshuck wants to ABX 2L's files.
"Some 2L Hi-Res music samples are either botched or cooked"
https://hydrogenaud.io/index.php/topic,111416.0.html

Where might I find that DSD file to try my procedure on?

Is in error when the PCM samplerate is 88.2k?

What happens with the old and new algorithms when you take the samplerate up so that a good portion of the noise shaping is still present?

Back to the topic. Just in case shakeshuck wants to ABX 2L's files.
"Some 2L Hi-Res music samples are either botched or cooked"
https://hydrogenaud.io/index.php/topic,111416.0.html

..., the group delay caused by the FIR filter used in the conversion may result in the two files not being exactly time-aligned, so care would need to be taken to ensure that they are time-aligned properly,

Quote from: bennetng on 2017-04-25 13:14:09

Back to the topic. Just in case shakeshuck wants to ABX 2L's files.
"Some 2L Hi-Res music samples are either botched or cooked"
https://hydrogenaud.io/index.php/topic,111416.0.html

Just getting back to the topic of "Some 2L Hi-Res music samples are either botched or cooked"  - from the few examples I can see in that thread, it seems to be a case of the exact phenomenon I was referring to earlier in this thread:

Quote

..., the group delay caused by the FIR filter used in the conversion may result in the two files not being exactly time-aligned, so care would need to be taken to ensure that they are time-aligned properly,

Using a linear-phase FIR for the Lowpass Filter in the conversion process (which is what is usually used) will inevitably cause a delay, and the delay time will vary depending on the length of the FIR filter used.
The length of the FIR (and therefore the delay time) will vary from one converter to another, and even within the same converter, when the LPF parameters are varied (eg cutoff frequency / transition band / steepness etc).

Therefore, unless I'm mistaken (and always happy to be wrong),  I don't consider slight timing variations in various conversions of the same original material to be evidence of "botching" or "cooking". It is just an inevitable consequence of having a linear-phase FIR filter for anti-aliasing.

Of course, you can avoid the delay by using a minimum-phase FIR, but then you will be introducing phase distortion, and since you will have changed the phase of individual frequency components, your subtraction trick will not work properly.

That's some extensive testing, tedsmith.
I'm happy to report that Peter seems to finally be adding ultrasound filtering to the foobar2000's ReplayGain scanner. Should improve ABX situation and perhaps even help any real RG user.

I think these DSD discussions often get confused because IMO there are two different questions and sometimes we end up with a moving target, deliberate or not:

1. CD quality vs. higher res, be it PCM or DSD.
2. DSD vs. higher res. PCM.

#1 is easier to test and if one can't ABX a difference, then #2 is pretty much academic.

Using a linear-phase FIR for the Lowpass Filter in the conversion process (which is what is usually used) will inevitably cause a delay, and the delay time will vary depending on the length of the FIR filter used.
The length of the FIR (and therefore the delay time) will vary from one converter to another, and even within the same converter, when the LPF parameters are varied (eg cutoff frequency / transition band / steepness etc).

Therefore, unless I'm mistaken (and always happy to be wrong),  I don't consider slight timing variations in various conversions of the same original material to be evidence of "botching" or "cooking". It is just an inevitable consequence of having a linear-phase FIR filter for anti-aliasing.

Of course, you can avoid the delay by using a minimum-phase FIR, but then you will be introducing phase distortion, and since you will have changed the phase of individual frequency components, your subtraction trick will not work properly.

Quote from: judd on 2017-04-26 12:45:50

Using a linear-phase FIR for the Lowpass Filter in the conversion process (which is what is usually used) will inevitably cause a delay, and the delay time will vary depending on the length of the FIR filter used.
The length of the FIR (and therefore the delay time) will vary from one converter to another, and even within the same converter, when the LPF parameters are varied (eg cutoff frequency / transition band / steepness etc).

Therefore, unless I'm mistaken (and always happy to be wrong),  I don't consider slight timing variations in various conversions of the same original material to be evidence of "botching" or "cooking". It is just an inevitable consequence of having a linear-phase FIR filter for anti-aliasing.

Of course, you can avoid the delay by using a minimum-phase FIR, but then you will be introducing phase distortion, and since you will have changed the phase of individual frequency components, your subtraction trick will not work properly.

I am not familiar with the relationship between filtering and delay, but from what I interpret from your reply, does it mean using a linear phase filter, in general, will introduce such a large amount of time delay?

Since the ABX log from mzil contains SHA checksums and 2L is still providing those files, I downloaded the 2496 file and use SoX to to convert them to 44k. I used the default settings as shown in the screenshot.

I inspected the converted file. As you can see in the screenshots, timing differences are basically nonexistent. (self-convert.png vs 96k.png)

Now, look at 2L's 44k file, the timing differences are so huge that I need to use a different zoom scale to show the differences. (2L 44k zoomout.png vs 2L 96k zoomout.png)

Since you also write your own converter so it should be fair to use your converter as an example. Indeed, your converter does produced some offsets, but the differences is much much smaller than 2L's and I would say negligible in ABX tests. (ferocious 44k zoomout.png)

I made SoundFonts in the past and used resamplers frequently, I did know some sample loops gone bad after resampled and I needed to re-align the loop point, but I never see a resampler can produce such a large offset.

Don't know if mzil is still visiting this forum or not...

The delay depends on the length of the filter.  For a steep linear phase filter, the length will be very long, and therefore so will the delay.   For a filter with gradual roll off, the delay can be very short, but the rejection will not be good.  If you don't need good rejection, you can therefore have a relatively low delay (or else you must be willing to give up being linear phase).

This is why I said before that if you want to compare DSD and PCM you have to look carefully at time alignment.  Filters that are likely to be transparent (good image rejection, linear phase) are likely to have long delays 

David's equal loudness filter adequately removes the noise shaping, whereas the hastily adopted (yes I did in fact just say that once again) shiny new little darling pet 1770 doesn't.

The point I don't understand is the SoX plugin can be easily downloaded and checked and by default (see the screenshot setting) it can already give a super beautiful sweep, why there is no delay?