Should I record at 48khz and convert to 44.1 for CD

  • Thread starter Thread starter Paul Ertel
  • Start date Start date
RE: You'll notice that the sampling rates mentioned were 88.2k and 96k-- which are exact doubles of 44.1 and 48 respectively. As I implied in my original post, exact multiples are not so bad when resampling. I didn't see anything there that recommends 48k over 44.1 and then downsampling, however.
 
i spoke with an employee of aardvark today and i asked him what he thought about whether or not you should record at 48khz or even 96khz to just convert back to 44.1khz. he said it really depended all on the project or type of music that was being recorded. but i feel he was basically saying just record at 44khz.
-Tom
 
OK- Here's my "Yahoo!"
I think it's a mistake to track at 48KHz and then dither down to 44.1 if you don't plan on doing any signal processing to the file. That is- if you're just gonna track and burn, leave it at 44.1.
The extra digital information comes into play when you process the file (EQ/COMP/NORM/sonic decimator etc.)
And I don't believe that 48 vs. 44.1 makes as much of a difference as 24 bits vs. 16. when it comes to applying digital processing.
The point pglewis was making is that a direct multiple downsample is much easier to program than a 48 to 44.1 conversion. Thus more likely that SW other than Pro level stuff will work fine.
 
Having dutifully searched before starting a new thread....

........I see that this is THE sampling rate debate thread. But I'm not sure whether to apply Ed et al.s' advice to my DIY context. I'm using an Aardvark Direct Pro 24/96 and N-track software. I'm recording acoustic singer/songwriter and blues stuff (let's assume for the sake of argument that it's fucking brilliant). I'm posting mp3's and making demos and eventually a CD. I want it to sound as good as possible and I believe that people respond to slight differences in quality even if they don't realize it. I think that N-track has some sort of limitation at 96k, but I can record and mix at 88.2k or at least something else much higher than 44.1k.

N-track can do the sampling rate conversion. So do I go ahead, record, mix, etc. to my heart's content at the higher rate, and then convert to 44.1k after my final mix, just prior to mp3ing or burning?
 
I think you should A/B the material and see for yourself what sounds better in your software but I can tell you that after speaking to Bob Katz myself,
he suggested not to use 48 over 44.1 if you don't own a top notch Sample rate
converter.
As for 96 or 88.2 - I didn't discuss this with him.

Like I said before, I would A/B the difference.
 
I'll try A/B-ing it, although it won't be exact of course since performances are a little like fingerprints.

What makes a sample rate converter good/lousy?
 
how about this?

mix down to 15ips 1/2" 2 track.. then send that into some 24/96 converters for mastering. another good alternative to that conversion question.
 
but then you get another d/a conversion!...unless you mean... GOD NO! you dont mean to track to tape initially!! NOOOO!

S A D

Simultaneous Analog Digital

xoxo
 
2 cents from a comp. sci guy

i addition to being a software architect for ibm, i am a professor of computer science at bowie state university, and have been involved in science technology projects for NASA, NOAA, and the Chinese Acadamy of Sciences dealing with real-time satellite systems. the issue is the real time ability of the equipment you use, and the algorithms that equipment employs. if you are using a software conversion algorithm offline, then your chances of success are enhanced. even when dealing with multi-billion dollar satellite systems that are supposed to be able to handle any circumstance, you have data dropouts. do you really think that won't happen with a piece of equipment that only cost a few thousand?

every second your DSP takes 48,000 snapshots of your audio wherein the value is between 0..16777215, and it must convert that value to 44,100 samples per second wherein the value is 0..65535 for your CD.

so let's immediately squash the idea of recording at 16bits because 24 bits is 256 times more accurate when dealing with the variation of sound pressure between each instance of a recorded sound. even if you have to dither in the end, 256 times more accuracy when throwing the tracks into some processing algorithm is better than the one time loss of dropping the remainder in the conversion from 24 to 16 bits.

now let's take the samples per second. 48000 / 44100 = 1.088435 (reduced for brevity). that means for each 1.088435 samples, you keep 1 sample. how is that possible? you don't take the sample unless the difference between the number of samples is > 1.088435.

here's an oversimplified example:
let's say you want to convert 8 samples to 5.

sample 1 (1 - 1.6 = -0.6 so throw it out)
sample 2 (2 - 1.6 = .4 keep it and hold the remainder)
sample 3 (1.4 - 1.6 = -.2 so throw it out)
sample 4 (2.4 - 1.6 = .8 keep it & hold the remainder)
sample 5 (1.8 - 1.6 = .2 keep it & hold the remainder)
sample 6 (1.2 - 1.6 = -4)
sample 7 (2.2 - 1.6 = .6 keep it)
sample 8 (1.6 - 1.6 = 0 keep it)

see how you end up with samples 2,4,5,7 & 8? not exactly evenly spaced is it? but remember we are talking about removing 3900 samples over a pool of 48,000 samples per second. that basicly means removing 1 sample for every 12 or so samples. synopsis: if you have an algorithm that solves the time sychronization issues demonstrated in the one above, you have no reason not to record at the higher sampling rate. i used a similar algorithm for displaying GOES-9 satellite images.

final thought:
a non-real time software algorithm for sample rate conversion is the best solution. if you are using a real time DSP, you take a calculated risk. but that risk is mitigated over time, because algorithms get better and hardware gets faster.
 
By necessity I've done a bit of experimenting with SRC. I started a project at 96K. I found that I needed to change to 48K, in order to get the number of tracks that I wanted, the system was just to stressed at 96K. I use Vegas so it was simple, Vegas does SRC automatically in playback, the original files remain at it's native rate. So, in mixdown the SRC is not done in real time- this does make a difference as crosstudio states above. I could not tell much difference between 96K and 48K (even in realtime), but at 44.1K there was some very noticable effects. Here's where it gets intresting, in order for me to run all the plug-ins I wanted, I "mixed down" the drums with the effects. The drums were recorded at 96K, the DSP applied to the drum tracks were right on. I sampled them to 48K in the process. The new, effected (compressed, EQ'd) tracks sounded "better" than the original files with the plug-ins in real time! (Ed knows what I mean, he said he's head this himself) Since I had such great success with it I wanted to "de-ess" this one vocal track the same way, to apply the DSP offline. The track was recorded at 48K, and when I appled the effects and played back the new track, it was horrible compared to the original track! In every situation I've come across a 96k file processed and resmpled to 48K sounded better than a 48K file processed and NOT resampled! And so now that the project is nearing completion, I will agian have to convert the sampling rate once I get to CD! Yahoooo!!!!! :D

-jhe
 
Yeah, but resampling from 96k to 48k is a trivial task.
We want to know what the difference is between sampling at 44.1k and sampling at 48k and resampling. If you have the time to do that (yeah I know it's a pain) that would be very interesting.
 
Thanks, James. If I have to end up at 44.1 (I have to, right, for a CD or to send it to mp3.com?), does it make sense to go right from 96 to 44.1 instead of going to 48 along the way?
 
It should (at least in theory) be easier to go from 96 to to 44.1 than from 48 to 44.1, since 96 has twice the amount of information, and more than double the info of 44.1. So yes, it makes sense. In fact, unless you are restricted by performace, it makes no sense to go via 48k. See the earlier posts on sticking with the original format as long as possible.

This I think everybody agrees on, right? I'm not starting a new flamewar here am I? :)
 
Yes, valid points by all. On most of the tracks, I am remaining at 96K, but on a few, I felt it would be better to go to 48K for more CPU headroom. (or actually I was stuck with it at 48K when I cut the vocals at 48K, forgetting to switch back to 96 in properties!! :eek: I damn sure ain't going to upsample my vocal tracks!) Going from 96K to 44.1K produced not so good results, thus went to 48K.
I feel that I am much better off resampling to 48K for those mixes (besides some of the tracks are 48K anyway), mastering with Wavelab, and then dithering and resampling to 16bit 44.1K for CD. Dither is very important and you can lose a lot at that step, I would like to have more samples to work with there. Even if it means only 10% more samples and yet another resample! I'm going to upload this one mix in question that has gone through many resampling steps so that you all can hear how it turned out. I'll post it as soon as I can upload it. (an all night venture on this line :()
Once I get a little more power in this system, I would like to start recording at 88.2K, to get that easy 1/2 SRC. Which I can't hear any difference in. I would be nice to know what the industry is going to do as far as getting a standard sample rate. DVD, of course, being 24/96. But what if for audio they start doing CD's in 24/88.2 on even just 16/88.2?? at this point it could be anything!

-jhe
 
Back
Top