Recording at 44.1 or 48

I'm curious, if given just those two options, which rate people are recording at?  I am familiar with both sides of the debate, but I've personally never had an issue with AA's ability to convert from 48 to 44.1, so I generally record at 48.  On the other hand, I can't say that in a "blind" test I could tell the difference, but do I like the idea of processing at the higher sample rate.

What is the rest of the world doing?

Thanks,
Sean

Edited to add: all of my material ends up at 44.1

The reason 48k is available is because DVD audio uses this rate. Most audio is 44.1k but higher rates can be used although it's just a waste on system resources.

higher rates can be used although it's just a waste on system resources.

There are those that would argue this point with you , but I'm not one of them! I probably would use higher rates if my system would handle them.

So I take it that you record at 44.1?

The difference between 44.1k and 48k is verging on the academic - unless you are a bat. Higher-still bitrate recordings are somewhat over-rated, too - even more so. I have systems that can certainly record stuff that only bats could hear, but since I have very few microphones that can respond to frequencies much above 24kHz, and virtually no home replay systems get anywhere even close to this, there doesn't seem to be a lot of point in using them at those rates. I did some comparison tests using the same feed to systems running at 48k and 96k, and nobody listening to the results on anything like normal speakers could identify a shred of difference between them.

Where there is a noticeable difference is in the bit depth you use; and even that difference seems to be primarily caused by converters apparently preferring  to run as 24-bit ones, rather than 16-bit. With all forms of recording, the quality of the mics you use, and possibly more importantly what you do with them makes far more difference than using esoteric bit rates. With the test recordings, if I moved the mics even a small distance, the results were immediately audible to most listeners.

There are A-D converters around* that sound better at 16-bit than most do at 24 - so there are other things that can affect the results too, like low dither clocks and decent analog stages.

You have to bear in mind that even 44.1k recording is going to give you a frequency response that extends beyond anything that you can hear by normal means, even as a young child (they always have the most extended HF hearing). Most, if not all of the claims made for higher bitrate recording are made either by marketing people, or somebody trying to justify a dubious wallet decision. Yes, we are now all forced to have all these systems, and yes, we could choose to use them at bat-like rates if we wanted to - but at the end of the day, if the punters can't tell the difference (and you probably won't really be able to either, if you are honest with yourself), what's the point?

So, I'd say that the chances are that a 96k 16-bit recording won't sound quite as good as a 44.1k 24-bit one, unless you use an Apogee*, or whatever, to do your 16-bit recording, when it probably will sound as good. But without the Apogee, it probably won't - the depth is more important, generally, in terms of a good listener experience. Are there any other exceptions? Well, with some older converters, the anti-alias filtering had effects that extended back down into the audible band if you ran them at 44.1k or 48k, but because the techniques used in modern converters use a somewhat different approach to antialiasing using oversampling techniques, this generally isn't an issue any more. And I think that it was some of these earlier converters and their awful brick-wall filters that were generally responsible for getting 44.1k and 48k a rather worse name than they deserved - it wasn't inherent, it was just poor implementation.

Steve,

I completely agree,  and I do record everythig at 24 bit depth (no Apogee here, just a lowly Echo Layla  ), but the issue of a software's ability to convert a file from 48 to 44.1 was a point of contention for quite a while - many feeling that the math required introduced a significant loss, and that was the (somewhat obtuse) point of my original post - If you are going to produce a finished product at 44.1, do you record at 44.1?

If I may inquire, what rate do you generally record at (I'm assuming the end product is generally 44.1)?

I always record at 44.1 (unless video is the destination, in which case I use 48k for better integration).  I cannot find any reason to record higher.  There is an argument that, in a top-flight studio, a good engineer could mix slightly more effectively at a higher sample-rate.  There is an equal argument that going from 48k to 44.1 will result in math errors.  In my mind, it's smart to record at 44.1 if the destination is 44.1.  But I doubt I could ever hear the difference either way.

I have seen some credible research that suggests that some people can hear a difference between 44.1 and 48k at playback.  However, as I recall, that same research suggested that, once downsampled to 44.1, no one could hear a difference.  I also recall that no one could hear a difference between 48k and 96k.  So it seems to me, at present, the sound quality of a 48k, 24-bit file is the highest quality anyone can hear.  Everyone I've ever seen that claims 96k sounds "so much better" has been one of those audiophiles that claim to be able to hear up to 35kHz and make other outragious claims.

Emmett

One point that should be borne in mind is that some cards (the Creative range, for example) only operate at one native sample rate - anything else going in or out is being resampled 'on the fly' - and that resampling is not of the highest quality.  So - if you are using a Soundblaster (as an example and against all the advice that has been given here) then you are better working at 48KHz (which is its native rate) and doing the final re-sampling in AA/CEP.

Otherwise, it's better to work at your target sample rate - in your case 44.1 KHz.  As has already been said, if it's quality and headroom you need, there's a lot more to be gained by upping the bit depth.

... but the issue of a software's ability to convert a file from 48 to 44.1 was a point of contention for quite a while - many feeling that the math required introduced a significant loss, and that was the (somewhat obtuse) point of my original post - If you are going to produce a finished product at 44.1, do you record at 44.1?

If I may inquire, what rate do you generally record at (I'm assuming the end product is generally 44.1)?

The math issue was only contentious with people who thought that it ought to be one - actually it isn't an issue at all if it's done correctly - which CEP and AA have always done better than the competition. To the extent that they are frequently used as sample rate converters for sample libraries - even some of the big ones, although their producers don't generally admit it.

What David Johnson realised was that sample rate conversion is about coding the direction and instantaneous rate of change of a waveform - and if you decode this information correctly, it's easy to recode it just as accurately at a different rate - with the proviso that if you encode it at a lower rate, and there are rate changes that occur faster than half of the new sample rate, you will distort them (aliasing) - so the CEP/AA converter applies post filtering to ensure absolutely no aliasing occurs.

And that is the only loss there is - if it's a loss at all, that is. It may well not be - certainly won't be if there's no content there. So with audio up to 20kHz, you won't hear a difference if you sample rate convert from 48k to 44.1k in Audition. This has been tried and tested so often now that it's actually not even slightly contentious.

There is an argument that, in a top-flight studio, a good engineer could mix slightly more effectively at a higher sample-rate.

I'd say that it was a statement, not an argument. There is no supporting argument with any legs that I'm aware of - it's simply untrue.

And differences heard between 48k and 44.1k will almost certainly be caused by in-band brick-wall filtering differences, or differences in performance of D-A converters - there's no hearing-related reason why anybody could differentiate otherwise. It's very possible that the filters could introduce quite severe phase differences at HF, and these most certainly could be audible, and in earlier devices they were. As I said earlier, this isn't generally an issue these days, and if you have equipment where it is, you should get rid of it!

My DAT machine records at 48k, but I hardly ever use it these days - so I don't have a choice about this. But generally, if the end result is going on CD, then I'll record at 44.1k. No names, no pack drill, but I had one client a while back who insisted on a 96k master, so I recorded everything at 48k, and sent him an up-converted result. Didn't get any complaints! And that's the point really - if you listen to a competent  recording in isolation, there is no way you can tell what the sample rate is, as long as it's at least 44.1k. So why record at a higher rate unless you are forced to by video cameras and DAT machines?

What I hadn't considered before was the 24 bit vs 32 bit side. Does this mean I am wasting space recording everything at 32 bit and would get just as good (bad?) results at 24 bit?

What I hadn't considered before was the 24 bit vs 32 bit side. Does this mean I am wasting space recording everything at 32 bit and would get just as good (bad?) results at 24 bit?

When recording, the ADC will not be providing more than 24 bits, so storing these as 32-bit floats is a waste of space.  However, once you start processing, the floating representation keeps more precision, so is preferable.  It's simpler just to stick with 32-bit unless you have a specific reason to do otherwise.

Paul

What I hadn't considered before was the 24 bit vs 32 bit side. Does this mean I am wasting space recording everything at 32 bit and would get just as good (bad?) results at 24 bit?

When recording, the ADC will not be providing more than 24 bits, so storing these as 32-bit floats is a waste of space.  However, once you start processing, the floating representation keeps more precision, so is preferable.  It's simpler just to stick with 32-bit unless you have a specific reason to do otherwise.

Paul

Whew! I thought for a minute I had been wasting all those giganytes for years!

The difference between 44.1k and 48k is verging on the academic - unless you are a bat.

I think that bats would not care for such difference either. I mean, if you comapred something recorded at 4.5 vs. 5 kHz sampling rate, you would be unhappy with each

I think that bats would not care for such difference either.

... unless you were a bat relying on a reflection at 23kHz! 

so I recorded everything at 48k, and sent him an up-converted result.

How do you know upconverting didn't add something to the mix? For my sound card that happens. I know we've had this discussion before.
I believe OVERTONES you DON'T HEAR afffect what you do hear. What happens in the bat zone affects what humans hear.

How do you know upconverting didn't add something to the mix?

A simple matter of spectral analysis - and if upsampling had added anything, it could only be described as noise and distortion - and of course it would have been inaudible.

For my sound card that happens. I know we've had this discussion before.
I believe OVERTONES you DON'T HEAR afffect what you do hear. What happens in the bat zone affects what humans hear.

a) How does his soundcard affect the file I send him? Well, it could do this, for all I know or care. But there again, so will every other component in his system, because it's not the same as mine.  And, b) I knew that he wanted to turn the results into a 44.1k CD, so what was the point anyway?

And you can believe what you like, for all the good it will ever do you. Doesn't alter the basic laws of physics, though... which say that if you can hear any difference, it's because of an in-band effect, not an out-of-band one. The definition of 'hear', incidentally, involves the concept of 'perceive', and since we know what the physical limits on this are, and they don't in adults generally get past about 17 or 18kHz on a really good day (usually less), I'd say that any arguments about the differences between 48 and 96k sampling are spurious - just like any differences would be. And the idea of 'don't hear' affecting 'hear' is a straightforward contradiction in terms.

If I played you a track here in isolation, and asked you what the sample rate was, you simply couldn't tell me - so that would be the end of the argument as far as I'm concerned - any relative differences are completely irrelevant. It either sounds okay, or it doesn't. And if it doesn't, this absolutely inevitably won't be anything at all  to do with the sampling rate,  for heaven's sake...