Downsampling from 88.2KHz vs 96KHz

[licendrivr]

It has been said that when recording audio for an audio CD, that 88.2KHz will be easier for a DAW to downsample to 44.1KHz than 96KHz.

I was wondering if anyone has actually experienced this? Though it would be hard to see since you'd need to record the exact thing twice, at both sampling rates to really A/B it....

Any thoughts?

[syra]

I haven't done this myself but when I asked Bob Ludwig he said 88.2 is less math so....there you go...

[arimaka]

I read somewhere where Bob Catz said it you lost less with sample conversion than what you gained from recording at the higher sample rates. I understand that when you go from 88.2 to 44.1 it just looses a sample each two samples. It's all an illusion anyway... Digital Audio that is...

Manuel Jimenez

[jlotto]

Honestly I can't hear 96khz or 88.2khz. I can't even hear differences between 44.1 & 48. 24bit sounds a little better than 16 bit though

[heinz]

It's strictly about the downsampling alogorithm and how it acts on the data. 88.2 is obviously an exact 2:1 stepdown to 44.1.

96k is to 44.1 is 2.176870748299319727891156462585:1. Not as clean.

The easiest way to correlate it is think of audio like a digital picture. If you reduce the size of a digital picture exactly 2:1, the math just involves throwing away every other pixel and leaving the remainder. You get the crispest possible image in the reduction because it retains the maximum amount of original image information as possible during the stepdown.

But if you take a 960x960 pixel image, and reduce it to a 441x441 pixel image, the math has to interpolate the difference in the new adjacent pixels to figure out where to put the original colors (in order to approximate the original data).

The result is a slightly less accurate stepdown, but not by much. It all depends on the quality of the algorithm. A digital artist can see it, but Joe Sixpack usually can't.

Which gives you better sound? Hell if I know, I still do 44.1/24.

[jlotto]

Originally posted by heinz


Which gives you better sound? Hell if I know, I still do 44.1/24.

[thethrillfactor]

According to Nika, everything is upsampled to the same rate(except 44.1khz) and then downsampled(for the conversion) to the final destination(44.1Khz), so it doesn't really matter what rate it is.

[arimaka]

Originally posted by heinz
It's strictly about the downsampling alogorithm and how it acts on the data. 88.2 is obviously an exact 2:1 stepdown to 44.1.

96k is to 44.1 is 2.176870748299319727891156462585:1. Not as clean.

The easiest way to correlate it is think of audio like a digital picture. If you reduce the size of a digital picture exactly 2:1, the math just involves throwing away every other pixel and leaving the remainder. You get the crispest possible image in the reduction because it retains the maximum amount of original image information as possible during the stepdown.

But if you take a 960x960 pixel image, and reduce it to a 441x441 pixel image, the math has to interpolate the difference in the new adjacent pixels to figure out where to put the original colors (in order to approximate the original data).

The result is a slightly less accurate stepdown, but not by much. It all depends on the quality of the algorithm. A digital artist can see it, but Joe Sixpack usually can't.

Which gives you better sound? Hell if I know, I still do 44.1/24.

I understand 100% what you are stating... But with 96khz you do capture a little more than 88.2khz.

From what I understood from Bob Catz... With old sample rate converters it was a problem because they weren't accurate enough. Now with newer sample rate converters you simply take a better picture with 96khz and then convert. I believe that's what most mastering engineer's do... Brad??? maybe you can come in on this one...

Mastering Engineer's take it back to analog and process the signal. Then bring it to digital on a great A/D at high sample rates and then downsample... right???

I'm just trying to pitch in...

[thethrillfactor]

Originally posted by arimaka
I understand 100% what you are stating... But with 96khz you do capture a little more than 88.2khz.

From what I understood from Bob Catz... With old sample rate converters it was a problem because they weren't accurate enough. Now with newer sample rate converters you simply take a better picture with 96khz and then convert. I believe that's what most mastering engineer's do... Brad??? maybe you can come in on this one...

Mastering Engineer's take it back to analog and process the signal. Then bring it to digital on a great A/D at high sample rates and then downsample... right???

I'm just trying to pitch in...

Yeah, but for certain types of music 88.2khz is a tighter sound then 96khz(which is at times too loose). Its almost like the difference betweeen 44.1khz and 98khz.

[Bob Olhsson]

I think this is a dangerous area to generalize in. Perfect converters would most likely sound the same at 44.1, 48, 88.2, 96 and above because there is no theoretical reason for them not to.

Unfortunately, we are all stuck using a variety of less than perfect converters. My guess would be that in many situations, the advantage of higher sample rates ought to be in inverse proportion to converter quality. In the end, we all need to go with what actually sounds best using our own converters and this can be expected to vary all over the map with different combinations of gear. This is why I harp on monitoring so much.

[Jules]

HA!

What an excelent swerve from the old fox!





arimaka wrote - "I understand 100% what you are stating... But with 96khz you do capture a little more than 88.2khz.

From what I understood from Bob Catz... With old sample rate converters it was a problem because they weren't accurate enough. Now with newer sample rate converters you simply take a better picture with 96khz and then convert"


I gravitate towards this understanding. I too have heard / read on line (????) that modern SRC doesnt exibit these old "problems"..

I go for the best 'capture I can, and I can hear the benifit of 96k converted material that has been SRCed down to 44.1

That old 82.2 to 44.1 theory could be a digital myth nowadays....

[Brad Blackwood]

Most modern SRCs are async so it really doesn't matter.

But again, if your ME is using analog gear, the point is moot.

Just do what sounds best to you.

[heinz]

Originally posted by Bob Olhsson
My guess would be that in many situations, the advantage of higher sample rates ought to be in inverse proportion to converter quality.

Yes, I also read this in Mr. Katz's book (hey! you have a quote callout in the chapter as well). The desire for higher sample rates initially came about because the filters were smoother/more accureate and yielded superior results.

The quote I think is dead on, "sonic differences have come down to mathematics in this new digital world".

[Jules]

"async"

wot is that?

[Brad Blackwood]

Asyncronous. Async allows 96k to 44.1k, while a syncronous SRC will only do multiples (96k to 48k).

Some say that sync SRC sounds better, but most tend to agree that good async is indistinguishable from sync.

[MMazurek]

The easiest way to correlate it is think of audio like a digital picture.

You have to be careful with digital photography analogies.

Going from 960x960 to exactly half, is NOT 480x480 (it would actually be one quarter the size).

Imagine a 2x2 pixel image (4 pixels like a box).

If they were like a checkerboard (black/white) and you need to 'downsample' to only 2 pixels worth of info, which would be black/white/etc...?

In real life, the image gets a little 'fuzzy'.

[Jules]

for Finalizer owners I thought you might find this link interesting...

http://tcsupport.custhelp.com/cgi-bi...cGFnZT0x&p_li=

I went for this upgrade on my Finalizer and use it mostly to SRC from

96/24 - 44.1/24

And also

96/24 - 44.1/16

And sometimes

48/24 - 44.1/16

It works a treat!

[heinz]

Originally posted by MMazurek
You have to be careful with digital photography analogies.

Actually the only real difference (as you pointed out) is that image data is downsampled on two axes, where audio data lies on a single axis.