Successful ABX testing to hear the difference between Redbook Audio vs upsampled to 192/24

[UltMusicSnob]

Quote:

OK, here is a set of new files, this time in FLAC format to reduce the download size. You can convert them to WAV first if you want to. The files are now attenuated by 0.63 dB to avoid clipping. I used the following command to simulate the "original" 192/24 file with attenuation:

 resample.exe -k 0.000002923047 -r 192000 -f 3 -ff 0.5 -fw 0 -fl -2400 -g 0.93 Test_File_Foobar_Redbook.wav test_192k.wav

I do not reveal the identity of all the files yet, so only listen, and do not analyze them.

However, B.flac is the 192/24 resampled file, so you can compare the others against that one.

From an earlier post, resample.exe is the program found at the link in your signature? If B.flac replaces the 192_24 I was originally using (yours with lower level to avoid clipping), then this adds another treatment difference, as my original was created with iZotope 64-bit SRC, using these settings: 32 db filter steepness, 500000 max filter length, 1.0 cutoff scaling, 175 alias suppression and 100% prering (from the Soundforge 10 conversion interface using the iZotope tool).
 
Okay, here's A vs. B. My subjective cue to listen for was soundstage depth/width.
This was a planned set of 20 rounds. I went very slowly here, lots of 10-30 sec silences between my 'play' clicks.

 
stv014, I read some of the content in the links in your signature, holy cr-- it's deep. Are you in the business as a professional, or just a really informed listener?

 

[Lenni]

for what is worth, they all sounded exactly the same to me.

 

[UltMusicSnob]

Quote:

for what is worth, they all sounded exactly the same to me.

You gotta be obsessed...

 

[stv014]

Quote:

 
From an earlier post, resample.exe is the program found at the link in your signature? If B.flac replaces the 192_24 I was originally using (yours with lower level to avoid clipping), then this adds another treatment difference, as my original was created with iZotope 64-bit SRC, using these settings: 32 db filter steepness, 500000 max filter length, 1.0 cutoff scaling, 175 alias suppression and 100% prering (from the Soundforge 10 conversion interface using the iZotope tool).

 
I do not use SoundForge, but I null tested my converted file against Test_File_Foobar_192_24.wav, and the difference is very small (peaks within +/- 1 16-bit LSB almost all the time, and most of that difference is above 20 kHz, as shown below; in other words, it is basically silence) without the attenuation. Of course, you can ABX that too with the "-g 0.93" option removed.
 

 

[stv014]

Quote:

Originally Posted by UltMusicSnob /img/forum/go_quote.gif
 
Okay, here's A vs. B. My subjective cue to listen for was soundstage depth/width.
This was a planned set of 20 rounds. I went very slowly here, lots of 10-30 sec silences between my 'play' clicks.

 
That looks impressive. How about the other files ?

 

[UltMusicSnob]

 Here's C versus B. Went to a different part of the file for this one, an isolated short vocal break.
 
 

 

[UltMusicSnob]

Quote:

 
I do not use SoundForge, but I null tested my converted file against Test_File_Foobar_192_24.wav, and the difference is very small (peaks within +/- 1 16-bit LSB almost all the time, and most of that difference is above 20 kHz, as shown below; in other words, it is basically silence) without the attenuation. Of course, you can ABX that too with the "-g 0.93" option removed.
 

I read through a couple of long discussions about filter steepness online, which seemed to be a place where folks discussing sample conversion placed a lot of their attention. In your description of the response.exe utility, that sounds like it corresponds to your lowpass filter window type, but I don't know what the '0' value means there. Is response.exe using a steeper or gentler slope than 32 db? iZotope maxes out at 200 db/octave, I think. The reason I ask is that some settings, they say, could create changes in the audible region (on a filter sweep test, I haven't seen it ABX-ed).

 

[UltMusicSnob]

D compared to B. I need to revisit this one. I couldn't tell you what the difference was, aside from holistic impression.

 

[stv014]

Only 2 more files to test in this set.

 

[stv014]

Quote:

 
I read through a couple of long discussions about filter steepness online, which seemed to be a place where folks discussing sample conversion placed a lot of their attention. In your description of the response.exe utility, that sounds like it corresponds to your lowpass filter window type, but I don't know what the '0' value means there. Is response.exe using a steeper or gentler slope than 32 db? iZotope maxes out at 200 db/octave, I think. The reason I ask is that some settings, they say, could create changes in the audible region (on a filter sweep test, I haven't seen it ABX-ed).

 
It is probably not the same window type, but I chose the parameters to make the output as similar to that of iZotope as possible. The graph I have shown is the average spectrum of the difference signal (0 dB would be the difference between a full scale sine wave and silence). In the audio band, it is mostly 24-bit quantization noise. In the 20 to 24 kHz range, which is the transition band of the filter, it is still only comparable to the level of 16-bit quantization noise.
 
Here is a graph comparing the filter frequency responses, and a zoomed in version:
   
The iZotope frequency response is extracted from music, so it is somewhat noisy, but it is enough to show that there is little difference. Both filters are linear phase as well.
Some might complain that there is "only" about 112 dB stopband rejection at 25 kHz (which rolls off to -140 dB at 34 kHz), but with other currently supported window types that would be needed to improve that (up to 200 dB), I cannot reproduce the filter response as accurately. Also, with the levels of high frequency components actually present in the signal, most imaging products are basically lost in the noise floor.

 

[UltMusicSnob]

Here's E vs B. A planned set of 20 comparisons.

 

[UltMusicSnob]

Replication of D vs. B. I couldn't get a really strong result on D yesterday (strong being <1% chance), so I wanted to prove to myself I could do it again, AND drive the mistakes way down.

 

[UltMusicSnob]

And, replication of E, again to prove to myself I could replicate it, at higher precision.
Also, listening at different portion of track, so that's good too.

 

[UltMusicSnob]

B versus F:
Having trouble "warming up" tonight.
I get the occasional 3/3 or even 4/4, but I can't keep it together for a planned run of 20, confidence always falls apart before I put together a run.
These "all in 192/24" trials are very tough, three may be my limit for a day. I'll try again tomorrow.

 

[stv014]

OK. Once you complete the testing, I will reveal what the files are. That is, assuming that someone else does not also want to try it.