The No.1 Website for Pro Audio
44,1 or 48 khz?
Old 14th May 2006
  #31
Quote:
Originally Posted by Bishbashbosh
Wow.
I don't think I agree with a single point of this.

Except that 48k is used for film/TV.....

Too bad.

His underlying theory is a little hazy but his basic advice is correct.
Old 14th May 2006
  #32
Quote:
Originally Posted by narco
Its a bit over simplified saying 44.1 output just discards every second sample. If you think about what happens in the A->D stage you'll realise that every second sample of 88.2 won't be the same as a 44.1 recording of the same source.

Even in 88.2 -> 44.1 every sample is going to be (should be) extrapolated under most situations

although overall I agree. Best solution is to record at least at 48 and master or mix in analog, or stay at 44 for all digital

narco
Geez.

No -- YOU think about what happens in the A/D stage...

Let's say you're simultaneously converting an analog wave form and you can start both converters at EXACTLY the same time. (Assuming that exactness is possible only for purposes of explanation, mind you.)

The first sample in the 88.2 file will have a time address of exactly half that of the first sample of the 44.1 file. The second sample of the 88.2 file will have exactly the SAME time address of the first sample of the 44.1 file. And so on...

With me so far?

Now... if the second sample in the 88.2 file is taken at exactly the same time as the first sample of the 44.1 file and everything else is equal... whaddya think those values are going to be? They're going to be the SAME.


So... why on earth would you want to extrapolate a different value?

It's ALREADY as acurate as it can ever be.


I mean... people... this is NOT brain surgery... just use your heads!
Old 15th May 2006
  #33
Lives for gear
 
John The Cut's Avatar
 

and without dragging Nyquist into the sample rate debate... the bits in between the sample points at 44.1 (ie. the bits you record with 88.2) are already accounted for by 44.1 (assuming you have a limited bandwidth, which you do.)

CDs sound alright don't they? Does anyone write a strongly worded letter to record companies complaining that they're missing out on half the bands music because its not recorded at 88.2?

Bit stupid that... but the hype speaks for itself.
Old 15th May 2006
  #34
There's for sure a lot of hype on higher resolution recording -- but I'm NOT saying there's no advantage in working at higher resolutions. (As long as they're an even multiple of the target rate, of course.)

Now, there's obviously no advantage (beyond having a high rez archival file) in recording something at 88.2 and then immediately bumping it down to 44.1 kHz. In that case, you really are in essence throwing away every other sample.

But if you're going to be mixing, processing, etc, your track(s), then, indeed, you should get some technical fidelity advantage to recording at the higher (even multiple) rate, doing your processing and mixing at that rate, and THEN doing a SRC on that. (Whether you hear the diff is another matter and will no doubt depend on a number of issues, not least of which is how much processing/mixing you were actually doing.)



PS -- To anyone I've offended with my exasperation and blunt talk today, please allow me to offer a general apology for my lack of patience and diplomacy. Mea culpa.
Old 15th May 2006
  #35
Here for the gear
 
DJ Spinn's Avatar
 

Smile

Quote:
Originally Posted by xabiton
very true. however at the end of the day consumers dont care as long as it sounds listenable and the lowest of the low end is still listenable to 90% of the audience. I think sometimes people look too much into quality which is why I think 16/48 is just fine for almost everything its beyond cd quality and most people arent lookin for anything more than that or can even tell for that matter

I totally agree - I record all my projects/tracks at 48kHz / 16bit. My masters get's tranfered to a DAT thereafter, and the songs sound as crisp, and clear as anything that I've previously recorded at 96kHz / 24 bit for experimentation.

Old 15th May 2006
  #36
Moderator
 
matt thomas's Avatar
Quote:
Originally Posted by theblue1
Geez.

No -- YOU think about what happens in the A/D stage...

Let's say you're simultaneously converting an analog wave form and you can start both converters at EXACTLY the same time. (Assuming that exactness is possible only for purposes of explanation, mind you.)

The first sample in the 88.2 file will have a time address of exactly half that of the first sample of the 44.1 file. The second sample of the 88.2 file will have exactly the SAME time address of the first sample of the 44.1 file. And so on...

With me so far?

Now... if the second sample in the 88.2 file is taken at exactly the same time as the first sample of the 44.1 file and everything else is equal... whaddya think those values are going to be? They're going to be the SAME.


So... why on earth would you want to extrapolate a different value?

It's ALREADY as acurate as it can ever be.


I mean... people... this is NOT brain surgery... just use your heads!
Ok - sorry but you're wrong

perhaps I should have mentioned the I have a Bachelor of Science from the University of Canterbury in Mathematics and Philosophy, all my graduate year being on digital coding theory, with specific reference to audio (and space craft communications), and no, I'm not making that up

its not that simple

common sence and a few magazine articles doesn't compete with a proper education

no more time for this argument
narco
Old 15th May 2006
  #37
Moderator
 
matt thomas's Avatar
ok, time for one more

I've just had a shower and felt guilty about being a bit rude myself in my last post, (ha I thought about you in the shower)

I can appreciate where you'r coming from, it would make sence to think that way.

As an illustration; I'll write this quickly as I have to go out, apologies for not being perfect.

Imagine in your model of a converter sampling a sine wave that was just over half of you your sampling frequency.

So at 44k we sample a 22.1 k sine wave. What happens under your model?

if the first sample will be on the null of the wave, the wave will have completed an entire half cycle by the time we get to the next sample, plus a little bit, so the sample will only be slightly off zero in amplitude. repeat as necessary

If we do this for every sample we will end up with a very low pitched wave in the bass register.

sampling the same wave at 88 k sampling rate we will end up with a very triangular wave at pretty much the desired frequency, as we also sample the peak of the wave between the first two samples you sampled at 44k

so at 88 k we will end up with a high pitched triangular wave.

Any body who writes a program to change 88 to 44 that converts this high pitched triangular wave to a low pitched sine wave will have produced a very "interesting" sounding algorithm

not that either wave would have nbeen sampled like this in the first place, this is just to illustrate what happens in the simplified view

in much the same way if you mak a vector based curve in adobe ilustrater then import it to photoshop you won't have a hard edge, also try changin the screen fonts setting on your computer for another illustration

hope that helps some

narco
Old 15th May 2006
  #38
Well, I'm just a dumb ass college dropout but -- damn it -- I hate to say it but I think that seems right to me.

I really hate to say it. heh

Obviously, I hadn't thought about frequencies that could be represented by the higher sampling rate but not the lower. (Nyquist, you ol' rascal, how could I forget you on that?)


Not sure a simple "my bad" is sufficient. How about mega mea culpa?


But -- as I think you indicated above -- the general advice to work at an even multiple of the target rate still holds, yeah?



PS... no apologies necessary. You weren't rude. You said "sorry" and I was wrong... And, honest to gosh, I DEEPLY appreciate being corrected. (And in a nice straightforward manner. Can you imagine how I would have handled the explanation? There'd be Venn diagrams and charts and paragraph after paragraph of tortured analogies and half-baked metaphors. Did I mention I was a college poet before I dropped out?)

____________


I think -- now that I've been humbled a little -- people are gonna like me more. What do you think?

heh
Old 15th May 2006
  #39
Lives for gear
 
Bishbashbosh's Avatar
 

Quote:
Originally Posted by gsilbers
i defenitly go for 88 or 44. just because you wanna have the least and less complicated math going on on your computer.but thats samples per second so for. but if its 88.2 (which is enough to sample those upper harmonics we will never hear but still need as a natural ocurrance and to avoid aliasing) then to dither it to 44.1 is just every sample in half while 96 to 44 is every sample divided into 2.18xxxxxxxxxx so thats a lot of extra change in the endng result that gets chopped of at 24 bits, and even more at 16 bit wordlegnth. if you work to make music for film/TV then yes use 96 and 48 cause thats the format its used.
No... sorry. Just wanted to go through this one point at a time:
Quote:
i defenitly go for 88 or 44. just because you wanna have the least and less complicated math going on on your computer
What?.... math is what computers do.
Quote:
88 its just half of 44 (duh)
Ok... that's probably just a typo

Quote:
a project doing 48+tracks 3-5 min each at 48 or 96 or 192 to later be dithered into 44.1 thats a hell a lot of math IDB that doenst mean ill do it 100% correclty and can create some imperfection on the audio quelity
I've NEVER seen any white paper, or ANY research that backs this up. For a start, you don't dither sample rate. Second..... well if your DAW can't cope with higher sample rates' more 'difficult math', then buy any one of the others that can.... The one thing that computers CAN do accurately is add things together. Exactly how does this affect audio quality?
That sounds like something somebody just made up.

Quote:
then to dither it to 44.1 is just every sample in half while 96 to 44 is every sample divided into 2.18xxxxxxxxxx so thats a lot of extra change in the endng result that gets chopped of at 24 bits, and even more at 16 bit wordlegnth
Yep... if that was how SRC worked (or PCM sampling for that matter).

Sorry, but this post is just audio myth regurgitated as fact.
Old 15th May 2006
  #40
It is not, however, a myth that sample rate conversion from a sample rate that is not an even multiple of the target induces alias error across the entire sample set -- by the very nature of the process.

Of course, alias error creeps in with most processing. But we try to minimize it.

If you have a choice for a full-digital-domain project, it makes sense to start with a rate that is an even multiple of the target rate to avoid 'extra' alias error. (Of course, that verbal shorthand includes the target rate, itself.)



[And I think my new friend Narco will back this basic advice up. :D ]
Old 16th May 2006
  #41
Lives for gear
 
John The Cut's Avatar
 

Quote:
Originally Posted by theblue1

I think -- now that I've been humbled a little -- people are gonna like me more. What do you think?
nah!
Old 16th May 2006
  #42
Lives for gear
 
John The Cut's Avatar
 

Well.. one good thing about never ending debates is that you always learn something new.

Now I admit I'm lost with Narco's explanation. You're talking about aliasing here right? We know wierdness is going to happen since your sample rate is less than twice the sampled frequency... in this example, i wouldnt expect a converter at 44.1 to capture a 22.1kHz signal accurately.

so a) thats what filters are for

and b) you're still talking about capturing a 22.1KHz sine wave! none of us can hear it so who cares!

We know that higher sample rates allow you capture higher frequencies. The debate is whether those frequencies matter or not.
Old 16th May 2006
  #43
Quote:
Originally Posted by mogWai
nah!
Yeah... probably right. I'll go back to being an argumentative jerk. I guess it's my dharma...


heh
Old 18th May 2006
  #44
Lives for gear
 
mixerguy's Avatar
Quote:
Originally Posted by Albert
My opinion would be to choose the sample rate based on the delivery format. 44.1 if the tracks are intended for CD, and 48k if they are video post production. If youa re mixing analog then you could choose 48k as well.

thumbsup thumbsup thumbsup thumbsup thumbsup thumbsup
Old 18th May 2006
  #45
Lives for gear
 
mixerguy's Avatar
Quote:
Originally Posted by Bishbashbosh
No... sorry. Just wanted to go through this one point at a time:

What?.... math is what computers do............ (snip) ...

I've NEVER seen any white paper, or ANY research that backs this up. For a start, you don't dither sample rate. Second..... well if your DAW can't cope with higher sample rates' more 'difficult math', then buy any one of the others that can.... The one thing that computers CAN do accurately is add things together. Exactly how does this affect audio quality?
That sounds like something somebody just made up...... (snip).....
Sorry, but this post is just audio myth regurgitated as fact.

Dude - use your freakin ears!

do this:

1) take a nice sounding 48K 24 bit stereo mix (if your mixes suck - borrow someone elses)

2) in Protools or DP convert to 44.1

3) now listen to how the new sound file has collapsed, and sounds flatter, and much worse.

4) now run your original mix from one DAW thru an Apogee Rosetta 200 to do SRC from 48k to 44.1k

5) now A/B compare the ITB SRC to the Rosetta 200 SRC.

WHAT SOUNDS BETTER!?!?!

I have done this test, repeatedly, with levels exactly matched. and everything carefully done. The Apogee SRC WIPES THE FLOOR with the DP or the PT SRC. NO CONTEST.

Like I said - clean out yer frikkin ears!!!

Old 18th May 2006
  #46
Lives for gear
 
synthoid's Avatar
 

mixerguy,

I believe you, but there's no reason whatever that this can't be done as well or better by an ITB SRC algorithm. It says to me that the software sample rate converters are defective. Sample rate conversion is a good job to give to a high-quality audio editor like SoundForge.

Anyone else find this discussion too focused on the final sample rate? What about the difference between plug-ins running at higher or lower sample rates? How is software multiband EQ affected by sample rate, and what gives a better result?

If you were doing digital photography, would you always choose an image capture size that is an integer multiple of the size you are going to print? Or would you simple capture the highest res photo that you can? What if you are going to do a lot of transformations to the image before printing it? What if you are going to print it in multiple formats? I for one don't find the answers to these questions to be obvious, neither in the case of images nor in the case of sound.

-synthoid
Old 18th May 2006
  #47
I've been doing computer graphics even longer than computer audio (and I put together my first 8 channel DAW in '96).

On the graphics thing -- it might not be practical -- but, yeah, if you want the best possible image you really do want to do do your resizing to a target of which your source size is an even multiple.

In fact, when I'm reducing the size of an image where quality is critical I will crop the image to a pixel size that can be 'evenly' reduced.

ie, if I have a 503 x 406 image that I want to reduce to, say, roughly 20% its current size (and I can get away with it aesthetically) I will slice those 3 pixels off the width and those 6 pixels off the height (to 500 x 400) just so that it can be reduced 'evenly' to 125 x400. (And then I may make a 'final' crop on that resized image for aesthetic purposes.)

Obviously that's not always practical -- but it will produce the best results.



But, I agree that the analog route vs software SRC is an intriguing dilemna -- at least for those of us on the outside of the math looking in.

By incurring an extra layer of A/D and D/A, you're obviously buying more alias error with those processes, as well.

It's not readily apparent to me why that often (always in my none too recent testing) produces better results than software SRC (which is why I'm in the habit of saying you "may" achieve better results via the analog route -- and, of course, for people who are already mixing outside the box, it's not really 'extra' D/A-A/D, since they're already doing it. So, for them, the whole issue is moot.)

But, at any rate -- if you know what you want your target rate to be, it still makes sense to pick your production rate based on that. (Obviously, increasingly, we are faced with a devil's bargain where material may be released in both formats. And then, I guess, you just have to pick which one is most important.)


Now we wouldn't be having this issue at ALL if the [uncharitable characterization] at Sony/Phillips had gone with the already established, somewhat higher fidelity 48 kHz broadcast digital standard.

But no.... they wanted a disc you could get an hour onto... for all those CRUCIAL hour long albums... you know?

I'll admit, as a classical music fan, I do appreciate that you could then get a really long symphony on a disc. But there aren't all that many hour plus symphonies and it's still not long enough for an opera. (Although I'm not an opera fan, so that's moot to me.)

And -- at least at first -- some of the double pop albums they put on single CDs actually had to have material REMOVED to fit...


So, when this stuff vexes you -- just remember WHO to blame -- Sony and Phillips.
Old 18th May 2006
  #48
Lives for gear
 
mixerguy's Avatar
Quote:
Originally Posted by theblue1
.......(snip)....
But, I agree that the analog route vs software SRC is an intriguing dilemna -- at least for those of us on the outside of the math looking in.

By incurring an extra layer of A/D and D/A, you're obviously buying more alias error with those processes, as well. .......

NO NO NO.

I'm sorry if I wasnt clear.

I go digitally from one DAW into Rosetta 200 (say, 48k) - I then activate the onboard SRC and it spits out the new sample rate (say 44.1) out of all the other digital ports.

I then record this digitally into a second DAW.

No analog stage included.

and... if I'm going 24 bits to 16 bits.... I'll use the Rosetta 200 UV22HR dither.

this also smokes any plugin dither I have used in ProTools.

The end result is WAY better than doing it in ProTools or DP. WAY better.
Old 18th May 2006
  #49
Gotchya!

Sorry about that. So, basically saying that all SRC processes are not the same, which makes sense.
Old 19th May 2006
  #50
Here for the gear
 

In highest end converters only bit depth brings benefits for audio...and i'm talking about 48bit mastering converters (at 44.1 or 48kHz).
In 'normal' high end pro converters (think of mytek or similar) it depends on the ad chip (!!!) what *sounds* better. Use 96kHz if possible for mytek or lower end (sarcasm!) converters when they sound better with this! Or if you plan to use software capable of dealing with it at this rate without any conversion throughout.
Don't use the converters dither or SRC. Don't use the daw's dither or SRC (ok, sonars / myteks pow-r 3 is relatively good, but you can do better...)
if you want it to be best use weiss (ehem a bit expensive but best possible) OR r8brain!!! (FOR FREE) for doing any conversion and have a real benefit from the better quality recording. Even 48kHz to 44.1 can make a remarkable step up in quality compared to 44.1 only with a high quality src/dither algorithm used. SRC and Dither can be much better than your sequencers built in crap.
Don't forget: it depends on the chip used and the converters dsp where it sounds best, i.e. what it is working best with (intended or incidently). Never downsample in the converter itself (ok, mytek makes it quite good, but you can do better).
UV22/cubase src is definately not the way to do it compared even to the free r8brain version...
There were shootouts showing this more than once.
Look out for downsampling plugins and avoid avoid avoid.
you can even benefit from a DIRECT SRC after recording in >= 48kHz.
This is not simple math.
It is as well electronics, IC's, dsp's, algorithms.
The theoretical scientific and mathmatical approach is correct for the incredible 48bit converters but not for any standard converters. Here it comes to much more factors. At least the human ear's crappy non scientific/mathmatical approach to audio is a factor that can drive most of the described theory into absurdity.
We have to deal with real world hardware and software. Think about that. And listen...

Not to offense anyone.

Kind regards

Martin

(OK, i studied physics and work with electronis and am a good programmer as well. OK, i didn't build a spaceship communication system but i don't want to record something that sounds like a bad telephone, ya know? heh fuuck
couldn't resist, sorry...)
Old 19th May 2006
  #51
heh none taken

I KNOW the converters in my Fireface and ADI-8 AE are not Lavry or Prism material
And they do sound better at 96 Khz than at 48.
And the Sonorus converters I sold, but a good friend now uses dayly, and that are almost 10 years old still rock at 48.
But right now. Fireface does the job. I am looking for a firewire 2 channel highend solution though, because I'm going to get me a macbook, or minimac. for recording the master, at 16 bit 44.1 Khz. (for cd, or 24bit 96Khz vinyl)
So I hope the Lavry black AD will have firewire (I know, it's not going to happen)
Please mr. Lavry, make us a box to use with a laptop, please!!!

Anyway, the SRC in Logic (7) is now OK. Powr 2 3 and normal dithering are available.
I like 2, i feed it slightly bright, for the music I'm doing it makes it nice and warm..

For a classy mac algorythm look here: (also nice comparison)
http://www.audioease.com/Pages/Barba...a4SRCTest.html
Old 19th May 2006
  #52
Lives for gear
 
RedWallStudio's Avatar
 

If its going to end up on a CD, its going to eventually be at 44.1 anyhow..well, that is a known fact. Everything else is a variable and the quality will depend on the gear that is in the chain. With that said, and if it is a concern, you should do a session in 96, 88, 48 and 44.1 and then compare them side by side and decide for yourself. I honestly couldn't hear a damn bit of difference on a finished reference cd, so it was a no-brainer (faster, less storage space) for me to stick with 44.1.
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump