GoldenSounds passes apparently ABX test for DACs (NOT Really)

[Blumlein 88]

? The spike has 40 dB amplitude over the noise floor.

I think you are mis-reading this graph. This is displaying the difference in the spectrum. Mostly there is 0 db difference. There is a spike of difference and then one filter attenuates lower past 20 khz than the other. So that isn't a noise floor, but a noise floor difference. Swapping the two files in Deltawave gives you this.

Looking at the spectrum in the other window shows while the difference is large the actual level of the spike is still very low.

Here you see the difference in the matched spectrum, and where the spike in difference occurs it is nevertheless around or below -100 db.

 

[danadam]

Yes, but this is a difference not actual level. So roughly one filter is -100 db where that spike is and the other is -90 db. I doubt anyone can hear -90 db at 22 khz.

Here's what happens around 22050 Hz:

 

[Blumlein 88]

Here's what happens around 22050 Hz:
View attachment 366505

Same thing as Deltawave.

 

[Merkurio]

Leaving aside the technical details of the test, it seems to me that the title of the video leads to confusion and deception, fueling the same false mantras that have plagued the hobby for decades (just look at the comments on the video to see how it serves to validate the same hive mind thinking about detecting differences in conditions that clearly aren't ABX).

 

[ObjectiveSubjectivist]

The question is where is the dacs test?
I don't see any...

 

[Blumlein 88]

Listening to the difference file in Deltawave, you can begin to hear things with 60 db of gain, and with 90 db of gain can hear bass and transients with the rest of the music at low level. If you let Deltawave correct for phase and EQ you have nothing left other than hiss at low level with 90 db gain.

It is interesting. Listening to the files I thought the normal filter had a touch of momentary distortion now and again on transients. I ran Dwave with a high pass filter at 15 khz. Listening to the raw filtered files there is nearly nothing to be heard except for about twice in the high performance file. While the normal file had a regular crackling sounds. You hear this even more in the difference file of these two high passed files. I find it unlikely to hear while masked in the the whole musical spectrum other than I seemed to hear it. Interesting results. I think something is off about the normal filter.

 

[Mad_Economist]

Listening to the difference file in Deltawave, you can begin to hear things with 60 db of gain, and with 90 db of gain can hear bass and transients with the rest of the music at low level. If you let Deltawave correct for phase and EQ you have nothing left other than hiss at low level with 90 db gain.

It is interesting. Listening to the files I thought the normal filter had a touch of momentary distortion now and again on transients. I ran Dwave with a high pass filter at 15 khz. Listening to the raw filtered files there is nearly nothing to be heard except for about twice in the high performance file. While the normal file had a regular crackling sounds. You hear this even more in the difference file of these two high passed files. I find it unlikely to hear while masked in the the whole musical spectrum other than I seemed to hear it. Interesting results. I think something is off about the normal filter.

Sorry if I'm misunderstanding, but wouldn't 60dB of gain result in illegal values? The delta waveform has peaks of up -46~dBFS.

 

[Blumlein 88]

Sorry if I'm misunderstanding, but wouldn't 60dB of gain result in illegal values? The delta waveform has peaks of up -46~dBFS.

When listening to the difference file with 60 or more db of gain there were places of probable clipping, in between you could hear the music at lower levels with an emphasis on bass.

When listening to the files being compared themselves I didn't use any gain. When listening to the files high passed I only used about 20 db gain. So why would the high performance filter not have issues, and the normal filter did have issues. Again this was listening directly to the file only while high pass filtered so everything below 15 khz was gone.

 

[Mad_Economist]

When listening to the difference file with 60 or more db of gain there were places of probable clipping, in between you could hear the music at lower levels with an emphasis on bass.

When listening to the files being compared themselves I didn't use any gain. When listening to the files high passed I only used about 20 db gain. So why would the high performance filter not have issues, and the normal filter did have issues. Again this was listening directly to the file only while high pass filtered so everything below 15 khz was gone.

Ahh, I get you now. You're right, the null depth here isn't especially deep, though with a 15k highpass I'm seeing a maximum value in the area around -80dBFS for the delta waveform

I'm not hearing what you're hearing on the normal filter/test B file, but I'll have Cameron have a listen tomorrow, and we'll have a closer look at the spectra of the points of greatest difference. Looking at the delta of the FFTs, the net effect is <.1dB over the whole duration of the song, but obviously it will be more significant in a single instance. I will say that -80dBFS content seems very likely to me to be masked by the music, so I suppose we agree there

I wasn't involved in the HQPlayer stuff (and don't have it), so I'll have to wait on daylight in the UK to follow up there

 

[amirm]

Was he using some kind of exclusive mode to listen to 21 kHz? Was he using REW?

 

[Mad_Economist]

Was he using some kind of exclusive mode to listen to 21 kHz? Was he using REW?

Cameron reported subjective audibility playing a 21khz sine in REW - not 100% sure what sample rate he was using.

Successively, after some skepticism internally, he did an ABX and consistently differentiated 21k and 22k sine files (but not 22 and 23).

 

[amirm]

Cameron reported subjective audibility playing a 21khz sine in REW - not 100% sure what sample rate he was using.

I was wondering if the pipeline was bit exact. Without it, you can definitely get artifacts in lower frequencies which would be audible.

 

[amirm]

Successively, after some skepticism internally, he did an ABX and consistently differentiated 21k and 22k sine files (but not 22 and 23).

What was the sample rate here? 176 kHz?

 

[restorer-john]

Cameron reported subjective audibility playing a 21khz sine in REW - not 100% sure what sample rate he was using.

Successively, after some skepticism internally, he did an ABX and consistently differentiated 21k and 22k sine files (but not 22 and 23).

Yeah, but at what level? You turn up 20k and IMD/HD products in an HPA or headphones can be audible.

 

[Mad_Economist]

I was wondering if the pipeline was bit exact. Without it, you can definitely get artifacts in lower frequencies which would be audible.

A good question! For the ABX of the tones, it was the Holo May's ASIO driver, so I'd presume it's bit exact.

What was the sample rate here? 176 kHz?

For the ABX, 48khz - not 100% sure what sample rate he used for the initial bit in REW.

Yeah, but at what level? You turn up 20k and IMD products can be audible.

How would you get IMD products from a pure tone? Further, if it were a distortion product, given that distortion usually rises as a function of frequency, wouldn't we expect 22 and 23k to be differentiable as well?

I don't believe he substantially cranked it from his prior listening level, but the RMS level of the sine wave .wav files looks to be around -6dbFS, so the peak level would be approximately -3.

Edit: the .wav files and ABX log are in the public drive folder if anyone wants to have a look!

 

[diddley]

 

[JSmith]

Really, this is a total mess. He should re-do the video and explain that this is a test of two filters in HQPlayer. It has nothing to do with DACs or audible differences between them.

Another disingenuous video.


JSmith

 

[sam_adams]

The conditions of the the test are 'cooked':

- the ABX test he is conducting seems valid.
- he got 18/20 correct. That's a p-value of 0.02%.
- the reason why is because of different reconstruction filters introducing a slight rolloff > 20kHz
- Goldensound is 26 years old. He can hear 21kHz. He was also specifically using headphones which don't have a Harman curve which rolls off the top end so he can hear HF better.

Change the response of the headphones via EQ or use headphones that don't follow the Harman curve:

Boosting the HF response at 20KHz by probably +35dB? Even Helen Keller could hear that.

The original track:

Probably not original if a CD rip because it's 24-bit , but that's incidental. Note the area outlined in red.

The resampled track, 176KHz, with 'normal' filter:

Note the area outlined in red.

The 'high performance filter':

Note the area outlined in red.

So, boost the response of the headphones so the HF is ragingly loud and distorting terribly. Use a filter that emphasizes a particularly small band of frequencies. Listen to the two tracks endlessly so that you are familiar with the differences under these artificial conditions. Post video on YT entitled, "Proof that DACs can make a difference", for clicks and Lulz with your "blind" ABX test results that don't really involve two DACs. Profit!

Another disingenuous video.

That is an understatement. The kid's got a future . . .

 

[diddley]

The conditions of the the test are 'cooked':



Change the response of the headphones via EQ or use headphones that don't follow the Harman curve:

View attachment 366553

Boosting the HF response at 20KHz by probably +35dB? Even Helen Keller could hear that.

The original track:

View attachment 366554

Probably not original if a CD rip because it's 24-bit , but that's incidental. Note the area outlined in red.

The resampled track, 176KHz, with 'normal' filter:

View attachment 366555

Note the area outlined in red.

The 'high performance filter':

View attachment 366556

Note the area outlined in red.

So, boost the response of the headphones so the HF is ragingly loud and distorting terribly. Use a filter that emphasizes a particularly small band of frequencies. Listen to the two tracks endlessly so that you are familiar with the differences under these artificial conditions. Post video on YT entitled, "Proof that DACs can make a difference", for clicks and Lulz with your "blind" ABX test results that don't really involve two DACs. Profit!



That is an understatement. The kid's got a future . . .

And a past behind him, He just gotta have it right.
Well maybe next time..

 

[KSTR]

No it isn't. For that, you would have had to simulate those filters, the ones in the DAC, not just two options on SoX.

We can do that as well... but I don't expect any different results than what we already know: If there is anything that might easily explain perceived differences between DAC then it's the reconstruction filters.