[xnor]

Quote:

[size=x-small]Anyway, I do empathize with a lot of people in regards to the 192KHz and 24-bit DAC trend. Since the popularity of the Benchmark DAC-1 - which I owned for years with lots of components, including Qualia, K1000, OMEGA II, $5000 speakers, etc. - every manufacturer has gone with 24/192 upsampling, which in many cases, make 16-bit recordings sound worse, not the same but worse. There are exceptions, but generally I prefer NOS. My favorite DAC is the CIAudio VDA-2 because it doesn't upsample 16-bit signals and simply sounds more natural, without artificial enhancement. 16-bit can be pretty good after all.  [/size]

There are couple of things I want to ask, but let's start simple. Who says that DAC doesn't upsample and what do you mean by upsampling exactly? Why do you think a lossless conversion like from 16 to 24 bits is an artificial enhancement? What makes you think that a signal left-shifted by 8 bits sound less natural?

 

[spaark]

Quote:

However, the human ear can definitely distinguish dynamic range above the 16-bit threshold. The sonic effects are small, but remember that to true audiophiles, even a 5% difference is enormous. People like Monty probably consider 5% to be nothing --thus their dogmatic conclusions. 

If you turn the volume way up, then yes. I don't think anybody listens at such levels to be able to hear the noise floor of 16-bit audio. Edit: Sorry, I didn't realize you were talking about dynamic range. My reply to that is still the same, though. You'd have to be listening pretty loud to notice, and of course the audio would have to have a DR greater than 96 dB.

 

[sonitus mirus]

With modern dithering techniques, you would need to crank up the volume to that of a jet engine in all but the quietest of environments.  Besides, what music has a dynamic range approaching 96dB?  I know my pathetic AC/DC CDs are probably pushing 10dB max range.

 

[skamp]

So some software programmer is more credible than Bob Katz (whom Monty regards as THE authority in digital mastering) when it comes to digital audio? OK, sure.

Hell yes. He knows a hell of a lot more about digital audio than any recording or mastering "engineer".

However, the human ear can definitely distinguish dynamic range above the 16-bit threshold. The sonic effects are small, but remember that to true audiophiles, even a 5% difference is enormous. People like Monty probably consider 5% to be nothing --thus their dogmatic conclusions. 

I don't know what a percentage would mean in this context. Actually, the difference between 16 bit and 24 bit at audible volumes down to -54 dBFS is less than 0.1 dB; less than half a decibel at -66 dBFS; 1 dB at -73 dBFS and 2.5 dB at -80 dBFS, as I demonstrated in a previous post in this thread. The actual differences are so ridiculously small or so ridiculously quiet, good luck trying to ABX that.

 

[bigshot]

Was Bob Katz the engineer that worked with Sheffield Lab back in the direct to disk vinyl days? Sheffield Lab did great sounding records, but they put out a lot of self serving anti digital propaganda about the sampling rate being too low compared to vinyl. They ended up eating those words and releasing a line of CDs.

 

[wuwhere]

Quote:

Was Bob Katz the engineer that worked with Sheffield Lab back in the direct to disk vinyl days? Sheffield Lab did great sounding records, but they put out a lot of self serving anti digital propaganda about the sampling rate being too low compared to vinyl. They ended up eating those words and releasing a line of CDs.

 
It was Doug Sax.

 

[bigshot]

Katz-Sax. My brain is turning to jello!
 
There were a couple of Katz having Sax outside my bedroom window last night!

 

[spagetka]

Hi, I'd love to try. Something dynamic, musical, jazz/classical....

1. Patricia Barber - Companion (2000)  - #1The Beat Goes On xxx ~~~
2. Patricia Barber - Modern Cool (1998) - #4 Constantinople ~~~
3. Hugh Masekela - Hope (2004) - #Stimela 12
4. The Dave Brubeck Quartet - Time Out (1959/r2013) - #3 Take five *** ~~~
5. Anne-Sophie Mutter - Carmen-Fantasie (1993) - #6 Sarasate: Carmen Fantasy, Op.25: 1. Moderato
6. Norah Jones - Come Away with Me (2012) *** ~~~
7. David Chesky - Area 31 (2005) ~~~
8. Johnny Cash - American IV: the Man Comes Around (2002) ~~~
9. Jimmy Cobb Quartet - Jazz in the Key of Blue (2009) *** ~~~
10. Collegium musicum - Collegium Musicum - Speak, Memory (2010) ~~~
11. Al Di Meola, John Mclaughlin, Paco De Lucia - Friday Night in San Francisco (1981) *** ~~~

 
*** I have different versions
~~~ - preferable
 
Thanks, M.
 
Quote:

 
Can you hear the difference ? Phasing or combing artifacts ? Try it with the foobar2000 ABX comparator:
 
c.flac
d.flac
 
One file is the 96/24 format original, and the other has been converted to 44.1/16, and then back to 96/24. The upsampling used a linear phase FIR filter that approximates what you would find in a decent real DAC.

 

[stv014]

Well, just upload a 30 second sample of whatever you think would be best for the test, and I will create the processed files.

 

[spagetka]

Hi, limit is 5MB which is 14s of 96/24 file
 
   
Thanks, have a good day.
 
 
Quote:

Well, just upload a 30 second sample of whatever you think would be best for the test, and I will create the processed files.

 

[stv014]

Here are a few filtered files you can try to ABX against each other and/or the original sample. They were created by downsampling the source to 44.1 kHz (using a long linear phase FIR filter with very fast roll-off), and then upsampling back to 96 kHz using various lowpass filters. Of course, these are only a few random examples of the infinite possible filter responses; perhaps I should also have included a filter with a very slow roll-off, and some loopback recordings from real DACs. Also, the sample is actually not ideal for this purpose, because it does not contain much high frequency information.
 
filters.zip
 
Frequency response (left) and group delay (right, 1 dB = 1 ms):
 
   
 
A_lp (red): no imaging, short impulse response, linear phase
A_mp (green): no imaging, short impulse response, minimum phase
B_lp (not shown): long impulse response (very fast roll-off), linear phase
B_mp (cyan): long impulse response (very fast roll-off), minimum phase
C_lp (yellow): no roll-off below 22 kHz, short impulse response, linear phase
C_mp (blue): no roll-off below 22 kHz, short impulse response, minimum phase

 

[jaddie]

Quote:

Was Bob Katz the engineer that worked with Sheffield Lab back in the direct to disk vinyl days? Sheffield Lab did great sounding records, but they put out a lot of self serving anti digital propaganda about the sampling rate being too low compared to vinyl. They ended up eating those words and releasing a line of CDs.

I still use tracks from the Lincoln Mayorga recordings for evaluating systems.  Admittedly, they've been digitized...sorry Sheffield/Doug.  I had the chance to use their plating facilities for a couple of projects, pretty good work. 

 

[spagetka]

[size=10pt]Thank you for the files. I agree that 14s sample file is not so good for expetiments. However, I will try to test it the week from 3rd March to 9th March.[/size]

[size=10pt]AFAIK FIR filter causes the oscillation. The thing is how many oscillations are "visible". For me FIR filter is something which can mask some things (and sometimes I like it ) but in the end I always got back to pure NOS DAC. But it is nice to have this possibility.[/size]

[size=10pt]I have played with up/down sampling in JRiver (DSP & output format) and the sound was slightly changed.[/size]

[size=10pt]In the past I have tested the files 16bit vs 24bit via Foobar ABX [/size]-> post #688
Br,M.
 
Quote:

Here are a few filtered files you can try to ABX against each other and/or the original sample. They were created by downsampling the source to 44.1 kHz (using a long linear phase FIR filter with very fast roll-off), and then upsampling back to 96 kHz using various lowpass filters. Of course, these are only a few random examples of the infinite possible filter responses; perhaps I should also have included a filter with a very slow roll-off, and some loopback recordings from real DACs. Also, the sample is actually not ideal for this purpose, because it does not contain much high frequency information.
 
filters.zip
 
Frequency response (left) and group delay (right, 1 dB = 1 ms):
 
   
 
A_lp (red): no imaging, short impulse response, linear phase
A_mp (green): no imaging, short impulse response, minimum phase
B_lp (not shown): long impulse response (very fast roll-off), linear phase
B_mp (cyan): long impulse response (very fast roll-off), minimum phase
C_lp (yellow): no roll-off below 22 kHz, short impulse response, linear phase
C_mp (blue): no roll-off below 22 kHz, short impulse response, minimum phase

 

[stv014]

Quote:

 
[size=10pt]Thank you for the files. I agree that 14s sample file is not so good for expetiments.[/size]

 
It is not because of the length that I think the samples are not ideal (if the sample is chosen right, even a few seconds can be enough to reveal a particular artifact), but rather because they are lacking the frequencies (> 10 kHz) where the filters would have the most chance to actually make a difference.
 
Quote:

 
[size=10pt]AFAIK FIR filter causes the oscillation. The thing is how many oscillations are "visible". For me FIR filter is something which can mask some things (and sometimes I like it ) but in the end I always got back to pure NOS DAC. But it is nice to have this possibility.[/size]

 
Well, the "B" filters with the very fast roll-off have rather long ringing, they use a 1 second impulse response. Can you hear it ? However, a NOS DAC will not eliminate ringing that is caused by the A/D conversion, or, in this particular case, the downsampling. For that, the best choice is in fact the "A" filter that has (almost) no imaging above 22.05 kHz, and the roll-off is as slow as possible while still keeping the attenuation at 20 kHz within 0.1 dB. The downsampler already used a 1 second long linear phase FIR filter that pre- and post-rings at 22.05 kHz. That can only be "dampened" to a shorter length if the upsampling filter has a very high attenuation already at 22.05 kHz, and its impulse response is short enough (shorter IR = slower roll-off).

 

[spagetka]

I have to say, it is pretty hard
 
test done (nb dell e6420 + AKG K701) - no external dac or amp

 
Quote:

 
It is not because of the length that I think the samples are not ideal (if the sample is chosen right, even a few seconds can be enough to reveal a particular artifact), but rather because they are lacking the frequencies (> 10 kHz) where the filters would have the most chance to actually make a difference.
 
 
Well, the "B" filters with the very fast roll-off have rather long ringing, they use a 1 second impulse response. Can you hear it ? However, a NOS DAC will not eliminate ringing that is caused by the A/D conversion, or, in this particular case, the downsampling. For that, the best choice is in fact the "A" filter that has (almost) no imaging above 22.05 kHz, and the roll-off is as slow as possible while still keeping the attenuation at 20 kHz within 0.1 dB. The downsampler already used a 1 second long linear phase FIR filter that pre- and post-rings at 22.05 kHz. That can only be "dampened" to a shorter length if the upsampling filter has a very high attenuation already at 22.05 kHz, and its impulse response is short enough (shorter IR = slower roll-off).