[EnOYiN]

Just downloaded the files and compared them (before looking at the spoiler for what that's worth.) and I couldn't tell the difference at all.

 

[xnor]

You're right.
 
The content above 22.05 kHz isn't really "recovered", it's just white noise shaped to look like the original content, hence I called it fake 96 kHz.
 
You don't have to use short lowpass filters, because anyone can do resampling from 44.1 to 96 kHz during playback, for example with the SoX plugin for fb2k. (I used the filter used in SoX for processing the files.)

 

[Kaffeemann]

Couldn't hear a difference.
 

 

[stv014]

Quote:

Originally Posted by xnor /img/forum/go_quote.gif
 
The content above 22.05 kHz isn't really "recovered", it's just white noise shaped to look like the original content, hence I called it fake 96 kHz.

 
Is the white noise shaped to look like the original, unfiltered 96 kHz signal, or is it generated entirely from the brick wall filtered version ? In the latter case, it is surprising how well the original content is "guessed" by the algorithm.
 
Quote:

Originally Posted by xnor /img/forum/go_quote.gif
 
You don't have to use short lowpass filters, because anyone can do resampling from 44.1 to 96 kHz during playback, for example with the SoX plugin for fb2k. (I used the filter used in SoX for processing the files.)

 
It is not that short, though, it still goes from less than 0.01 dB attenuation at 20500 Hz to more than 112 dB at 23000 Hz, but it is intentionally not like what I used for downsampling (as shown on the right channel below) so that it is more similar to real DAC filters, and is a reasonable compromise between flat response in the passband, stopband attenuation, and short ringing.
 

 

[xnor]

Quote:

Is the white noise shaped to look like the original, unfiltered 96 kHz signal, or is it generated entirely from the brick wall filtered version ? In the latter case, it is surprising how well the original content is "guessed" by the algorithm.

The FR was adjusted to look similar to the averaged spectrum of the original, but all the rest of the processing is based on the filtered version. There's a simple reason why I did this and I think you can guess why.
 
The algorithm is pretty simple, but could be changed to detect harmonics and dynamically boost the corresponding frequencies accordingly. I don't have time for that though and don't think it would make a difference. Even total absence of HF content shouldn't make a difference, except for those using their eyes instead of ears.

 

[parnold]

I was close! 

 

[xnor]

Quote:

I was close! 

Yeah, close guess but I was more interested if you could hear the difference.

 

[Averruncus]

Spoiler: Warning%3A%20Spoiler!

I'm surprised I managed to guess C as the original. Once again, more of a gut feeling then actually hearing any difference, I would probably do poorly in an ABX test (which I honestly don't think is a good test for audibility for very subtle differences).
Still, I did find B to be the worst, which makes me curious on how you filtered the freqs beyond 22KHz. When you resampled it to A, the resampler probably had a smoother filter Q and better aliasing compared to what you did in B, which I suspect made it sound better as the transients would sound more natural (bit depth probably made no difference). If not for B, I would probably not have been able to give a guess at all.
 
Edit:
 

Red line is the difference between C and B (inverted phase) while the yellow line is the difference between C and A.
Actually, it looks like shaving off the 48dB from 24 to 16 bit smoothed out the higher range frequencies, making it less distinguishable from the original, just with a bit more noise.

Interesting test. I'm glad my work setup (fubar IV via USB -> FA-003) was able to give me enough fidelity for a guess.

 

[PFULMTL]

Well at least I got the worst one right.  The other two are acceptable.
 
Do another one!  Once a week tests would be fun haha.

 

[wuwhere]

What is right or wrong? From the comments, it was preferential or interpretation of what we heard.

 

[bigshot]

You're choosing between a full range high bitrate master and a regular CD with noise above the audible spectrum. I think that pretty much shows that you don't need those super audible frequencies or high bitrates.

 

[wuwhere]

Quote:

You're choosing between a full range high bitrate master and a regular CD with noise above the audible spectrum. I think that pretty much shows that you don't need those super audible frequencies or high bitrates.

 
+5.

 

[bigshot]

I think the most interesting thing about this particular test is the clever way xnor made it so no one could peep at the file and figure out which was which by looking at the waveform. That shows how dedicated people are to proving their conclusion at all costs. They aren't interested in the truth. They want to cheat to make sure their argument isn't discredited.
 
This shows that there needs to be a put up or shut up to these tests. Taking people at their word isn't good enough. They need to verify that they aren't cheating.

 

[wuwhere]

That's the interesting part after xnor revealed the differences. I will always trust my own preference no matter what is right or wrong to somebody else. CJs are inaccurate but they are very pleasant to listen to. I don't listen to digital music to take it apart bit by bit. I listen to digital music because I enjoy the sum of the bits.

 

[Currawong]

I guessed (before I read more than the first 3 posts) it was going to be a 44.1 vs. 96k test. 
 
Before reading any other posts though, my first thought was C sounds most natural, as the trumpet was ever so slightly less harsh (but I had trouble believing myself after listening many times because the difference was so small), followed by A, which still has the sound of the musician shifting in his seat, followed by B which has something more akin to static instead of that sound. I had guessed that B had been compressed into an MP3 or similar first though.
 
Thanks for the interesting test.