Skip to main content

Topic: 96 vs. 48 or 44.1 kHz sampling --> scientific test (Read 66715 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • Bedeox
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #25
Quote
2. Downsample the recorded digital signal to 44.1 and 48 Khz afterwards on a PC, with a normal FFIR filter ( or whatever ), but using a very high internal precision to avoid rounding errors ( like 32 bit FP ), and dither the signal

This is a mistake... filtered sound is not same as original...
Anti-aliasing curve might not be perfect...
And 32bit floating point is not high internal precision.
Use 48, 64 or even 80bit for best results. (an overkill, but it won't hurt)

I think you should record three times:
a ) record 'reference signal' using highest possible quality with no filtering/dithering
b ) play it back analog and record as 44.1kHz 24bit, then dither.
c ) play it back analog and record as 48kHz 24bit, then dither.

Dithering and all records ought to be made using same setup.
I know this will add generation loss,
but it should be identical in both cases.

<edit>
Forum changes b ) to B)...
</edit>
  • Last Edit: 31 January, 2003, 03:34:42 AM by Bedeox
I've changed only because of myself.
Remember, when you quote me, you're quoting AstralStorm.
(read: this account is dead)

  • NumLOCK
  • [*][*][*][*][*]
  • Developer
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #26
Bedeox:  Why not record the analog to 96kHz, 24 bit, only once ? That would rule out any analog issues from the start.

The goal is to see if audible information gets destroyed by storage in lower sampling rates, right ?

So, a solid and reliable way to do it, could be:
a ) 96kHz/24bit is the reference signal.
b ) Brickwall lowpass the reference @ 23.5 kHz;  downsample to 48kHz/24bit with dither;  upsample to 96kHz/24bit with dither;  we have test signal #1.
c ) Brickwall lowpass the reference @ 21.5 kHz;  downsample to 44.1kHz/24bit with dither;  upsample to 96kHz/24bit with dither; we have test signal #2.
d ) Do the same as "b", but use 20-bit as downsampling target, then upsample back to 24. We have test signal #4.
e ) Do the same as "c", but use 20-bit as downsampling target, then upsample back to 24. We have test signal #5.

Now compare each of the 4 test signals against the reference signal.
Testing #1 and #2 will tell whether storing digital audio in 48kHz, 24-bit and 44kHz, 24-bit is harmless.
Testing #3 and #4 will tell the same about 20-bit.

Remarks:
- The same 96/24 DAC is used each time, in the exact same mode.
- Brickwall filtering should be easy in digital at that high sampling rate.
Try Leeloo Chat at http://leeloo.webhop.net

  • Bedeox
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #27
Lowpass filtering sound introduces differences... I have thought about it too.

<edit>
It SHOULD be unnoticeable, but you never know... these frequencies MIGHT have some impact
on lower ones...
</edit>
  • Last Edit: 31 January, 2003, 04:19:26 AM by Bedeox
I've changed only because of myself.
Remember, when you quote me, you're quoting AstralStorm.
(read: this account is dead)

  • NumLOCK
  • [*][*][*][*][*]
  • Developer
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #28
Quote
Lowpass filtering sound introduces differences... I have thought about it too.

<edit>
It SHOULD be unnoticeable, but you never know... these frequencies MIGHT have some impact
on lower ones...
</edit>

Well, don't worry, using a high-order FIR lowpass filter, the introduced errors will be way below the noise floor of the best 24-bit A/D converter. Remember, you don't even need the filter to be doable in realtime !

Anyway, the goal is to determine
- whether storage in 48kHz (and 44kHz) in audibly inferior or not, to storage in 96kHz
- whether storage in 20-bit and 16-bit is audibly as good as storage in 24-bit, or not.

For this purpose, you take a reference signal, and try to generate the best possible 44- and 48-kHz waves from it. The quality of the filter will matter a bit of course, but it is perfectly normal that the final result takes everything into account !

For example, if no difference can be heard when going from 96kHz/24bit to 44kHz/20bit, it will mean that:  even when taking ALL digital filtering imperfections into account, there is NO REASON to use more than 44kHz/20bit 

Then, if people want to store audio safely, they can be confident that a sampling rate of "X-kHz" at a bitdepth of "Y-bit", processed by the filtering algorithm "Z", will be enough for human ear  B)
Try Leeloo Chat at http://leeloo.webhop.net

  • Bedeox
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #29
Who wants to make such a test and has the equipment,
people and time to do it? Any volunteers?
I've changed only because of myself.
Remember, when you quote me, you're quoting AstralStorm.
(read: this account is dead)

  • Pio2001
  • [*][*][*][*][*]
  • Global Moderator
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #30
What about the consumer products ?

Even if 48 kHz 20 bit is enough in theory, wouldn't it be possible to design cheap 24/96 kHz converters that would play 24/96 audio with the same quality as an expensive 20/48 one ?

  • NumLOCK
  • [*][*][*][*][*]
  • Developer
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #31
Quote
What about the consumer products ?

Even if 48 kHz 20 bit is enough in theory, wouldn't it be possible to design cheap 24/96 kHz converters that would play 24/96 audio with the same quality as an expensive 20/48 one ?

Well, yes, we could feed the 24/96 a nicely upsampled wave, which would rival expensive 20/48 ones without problem 

By the way, couldn't we measure the typical linearity of D/A converters, and compensate almost perfectly, by table-lookup in software, just before sending the 24-bit sound to the card ? This would dramatically reduce THD, no ?
Try Leeloo Chat at http://leeloo.webhop.net

  • Bedeox
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #32
What is a 'typical' linearity?

Every type of DAC has other characteristic... do you want to measure them all?
And what about speakers and room setup? Or different types of headphones...
I've changed only because of myself.
Remember, when you quote me, you're quoting AstralStorm.
(read: this account is dead)

  • mithrandir
  • [*][*][*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #33
Arguing over the wrong things...

I would be perfectly satisfied with 44.1/16 for life if popular recording engineers actually cared about sound quality instead of producing highly compressed, overloud and clipped crap that gets shoved on the market today. New formats can't fix the engineer problem.

  • F1Sushi
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #34
Quote
Arguing over the wrong things...

I would be perfectly satisfied with 44.1/16 for life if popular recording engineers actually cared about sound quality instead of producing highly compressed, overloud and clipped crap that gets shoved on the market today. New formats can't fix the engineer problem.

I think someone just hit the nail on the head...


  • Bedeox
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #35
You're so much right, mithrandir... a lot of music nowadays is so badly compressed,
that it just isn't music anymore, because music consists of:

1. Rythm
2. Tempo
3. Melody
4. Dynamics (change in volume, musical term, not technical one)

So, this 'music' has next to no dynamics...
I've changed only because of myself.
Remember, when you quote me, you're quoting AstralStorm.
(read: this account is dead)

  • F1Sushi
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #36
Does anyone remember this article on compression and limiting? It is a sad but true tale about the current state of the recording industry...

http://www.prorec.com/prorec/articles.nsf/...6256C2E005DAF1C

  :x

  • budgie
  • [*][*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #37
Uh?
  • Last Edit: 01 February, 2003, 08:17:19 PM by budgie

  • F1Sushi
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #38
Quote
Quote
Mithrandir Posted on Jan 31 2003 - 07:40 AM
Arguing over the wrong things...

I would be perfectly satisfied with 44.1/16 for life if popular recording engineers actually cared about sound quality instead of producing highly compressed, overloud and clipped crap that gets shoved on the market today. New formats can't fix the engineer problem.


And if you know it already, then excuse me, I just felt like telling it from THE OTHER SIDE... Believe me, recording engineers are rarely to blame!

Now THAT'S a rant I can live with...


  • Bedeox
  • [*][*][*]
96 vs. 48 or 44.1 kHz sampling --> scientific test
Reply #39
Oh yeah... THE OTHER SIDE you say?
From INSIDE of the grave, you think?
(Budgie = disturber of peace!)

Well, we can blame the destruction of sound in most 'popular' tracks
on mastering engineers.
  • Last Edit: 31 January, 2003, 02:53:22 PM by Bedeox
I've changed only because of myself.
Remember, when you quote me, you're quoting AstralStorm.
(read: this account is dead)