Jump to content

Recording advantages of 44.1kHz vs. 48kHz revisited


UstadKhanAli

Recommended Posts

  • Members

My friend sitting here brought up a good point. We've been conditioned to record 44.1kHz due to CDs being 44.1kHz.

But since CDs are disappearing, and we use audio for videos more commonly, including uploading 'em to YouTube or playing back MP3s, is there any disadvantage in recording at 48kHz?

What are your thoughts?

Link to comment
Share on other sites

  • Replies 84
  • Created
  • Last Reply
  • Members

No disadvantage that I know of. I've been tracking at 48k since about 2004 or so.

Here's something that I can't figure out...

If you're ever planning on providing hd tracks to something resembling a "market", you better track at 96k or 192 .. like now.

I always figured I could port stuff or upsample to 96 on existing tracks but you know what?... there are guys in high fidelity land that know when you do that. They can tell from scopes and screens and things.

How do they know that? That really throws me. If I do a real-time port of a 48k 24bit song into a second daw at 96k... those guys ALWAYS know that the source was 48.

And then they call foul.

How do they do that?????? I'm going to have to find out more from those expert high-fidelity nerds.

Anyway, I'm probably going to ramble on over to 96k pretty soon anyway for tracking.

There's really no reason to avoid it now I guess.

Bought my first cell phone this year so anything's possible.

Link to comment
Share on other sites

  • Members

The only disadvantage to 48K is slightly bigger files, and a very slight (to me inaudible) degradation when converting to 44.1K for CD. The latter is minimized with good software; all are not alike.

Regarding folks being able to detect upsampling: Just a guess, but I bet if you upsampled all the individual tracks and then mixed at 88K, the upsampling would be obscured. Keep in mind that FX would run at ful 88K, so for any nontrivial FX, it wouldn't be the same as upsampling the result.

I have a friend who swears that doing this (recording at 48, upsampling to 88 before rendering) and then downsampling made the mix sound better, even though they weren't using any fancy FX, just a bit of reverbs and delays. He says it was a blind study and all the bandmates preferred the 88K mix. I'm skeptical of the claim, but he's a very intelligent and reliable guy. Of course, that's one data point and may not be significant.

I'm skeptical because upsampling adds noise, downsampling adds noise, and linear FX like delays and reverbs (especially convolutions, but maybe not all reverbs) are *linear* so there's no time-domain improvment in running at a higher rate. Nonlinear stuff like chorus or pitch shift doubling could very easily sound better at higher rates, especially if the algorithms aren't as careful as they should be about anti-aliasing.

Link to comment
Share on other sites

  • Members

I've been recording lately at 96/32-float. I like the sound--- it seems to me that vst/dx filters I apply exert less cumulative damage/noise to the audiofile at those rates. A 44/16 file seems to get awfully distorted by the time you've treated it with various dx/vst filters. I could be wrong. You guys understand these things much better than I.

Link to comment
Share on other sites

  • Members

I've also been recording at 32-bit float.

In light of what my friend suggested the other evening and what you guys are writing here, I think I'll probably switch back to 48kHz for any new projects.

Not sure about 96kHz yet. File size, etc. My system probably won't run so well, so I think that will have to wait.

Link to comment
Share on other sites

  • Members

Storing recorded tracks as 32-bit float wastes 1 byte per sample. There is no additional information in the file; your soundcard gets 20 bits of significant audio data, adds 4 bits of white noise (which is a good thing), and delivers a 24-bit sample. Converting it to 32 bits before storing to disk just wastes space and disk I/O. But if you're not hitting any space or disk bandwidth limitations, then fine. It saves a little CPU time on playback, since the samples don't need to be converted from fixed to float.

However, 32-bit float has a huge advantage for mixes that will have any further processing (e.g., mastering). In 32-bit float format, there is no clipping. If you go over 0 dB "FS" (sort of a misnomer, because it's full-scale only in fixed formats), there's no loss in precision, no lopping off of peaks, etc.

The only possible harm to going over 0dB in float format is that some DSP algorithms might not be optimized to handle values over 0dBFS well. (DSP algorithms involve a complex set of tradeoffs; in some cases the high-level matrix math looks simple, but in practice intermediates will hold near-infinitesimals or near-infinites, and so the algos have to be modified to avoid these. BTW, this is about the only reason why 64-bit float might sound better than 32-bit float.) But that's far better than clipping. The solution is the same in either case (turn down the gain somewhere, or use compression/limiting, or whatever your favorite technique is for taming rogue peaks. With a fixed format, you have to do it before mixdown. With float, you can fix it after mixdown, and if you're going to do it using master-channel methods, there's really no difference (between fixing it pre- vs post- mixdown).

Oh, the only other harm is that you might just play the clip, and it'll sound bad if there's much content over 0dBFS.

Link to comment
Share on other sites

  • Members
Quote Originally Posted by philbo View Post
You wouldn't need to go through tape (though you could, and it sounds nice). Just loop the outputs to the inputs with a cable. The filtering on the D/A output will fill in the gaps just fine.
That would be easy with two computers. It might also be possible with two DAWs that can run concurrently, and two soundcards.

But I don't know of a DAW that can record inputs at one rate while playing to outputs at a different rate. I also doubt that most soundcards can run different ports at different rates concurrently.

Plus you wouldn't get that warm tape compression. ;-)
Link to comment
Share on other sites

  • CMS Author

 

Quote Originally Posted by UstadKhanAli

View Post

I've also been recording at 32-bit float.

 

You understand that the 32-bit float(ing point) doesn't have anything to do with the sampling. The audio is still sampled to a word length of 24 bits. What's 32 bits is the arithmetic used for processing the data. Since the word length grows with just about every operation, using 32 bit floating point arithmetic prevents the audio data from being mangled by truncating or rounding to fit a 24-bit word.
Link to comment
Share on other sites

  • Members
Quote Originally Posted by MikeRivers View Post
You understand that the 32-bit float(ing point) doesn't have anything to do with the sampling. The audio is still sampled to a word length of 24 bits. What's 32 bits is the arithmetic used for processing the data. Since the word length grows with just about every operation, using 32 bit floating point arithmetic prevents the audio data from being mangled by truncating or rounding to fit a 24-bit word.

I believe Cubase samples at 32 float when set to do so..
Link to comment
Share on other sites

  • Members
Quote Originally Posted by Bookumdano2 View Post
I always figured I could port stuff or upsample to 96 on existing tracks but you know what?... there are guys in high fidelity land that know when you do that. They can tell from scopes and screens and things.

How do they know that? That really throws me.
I was thinking it was obvious--maybe they just notice the complete absence of frequencies over 24k. But I've never actually put a high sample rate recording on a frequency analyzer to see if there is any information there at all.

This blog is as good an article as any I've seen on the topic: http://people.xiph.org/~xiphmont/demo/neil-young.html
Link to comment
Share on other sites

  • Members
Quote Originally Posted by learjeff View Post
Storing recorded tracks as 32-bit float wastes 1 byte per sample. There is no additional information in the file; your soundcard gets 20 bits of significant audio data, adds 4 bits of white noise (which is a good thing), and delivers a 24-bit sample. Converting it to 32 bits before storing to disk just wastes space and disk I/O. But if you're not hitting any space or disk bandwidth limitations, then fine. It saves a little CPU time on playback, since the samples don't need to be converted from fixed to float.

However, 32-bit float has a huge advantage for mixes that will have any further processing (e.g., mastering). In 32-bit float format, there is no clipping. If you go over 0 dB "FS" (sort of a misnomer, because it's full-scale only in fixed formats), there's no loss in precision, no lopping off of peaks, etc.

The only possible harm to going over 0dB in float format is that some DSP algorithms might not be optimized to handle values over 0dBFS well. (DSP algorithms involve a complex set of tradeoffs; in some cases the high-level matrix math looks simple, but in practice intermediates will hold near-infinitesimals or near-infinites, and so the algos have to be modified to avoid these. BTW, this is about the only reason why 64-bit float might sound better than 32-bit float.) But that's far better than clipping. The solution is the same in either case (turn down the gain somewhere, or use compression/limiting, or whatever your favorite technique is for taming rogue peaks. With a fixed format, you have to do it before mixdown. With float, you can fix it after mixdown, and if you're going to do it using master-channel methods, there's really no difference (between fixing it pre- vs post- mixdown).

Oh, the only other harm is that you might just play the clip, and it'll sound bad if there's much content over 0dBFS.
If you don't have gain staging mastered, you don't need to be worrying about 32 bit floating point vs 24 bit fixed point. You need to be worrying about gain staging. My 2c.
Link to comment
Share on other sites

  • Members

FWIW of the past say 100 or so records I've mixed, unsigned, indie or major label, almost ALL of them have been 44.1kHz/24bit. I got one project that was 48kHz from an unsigned artist and one project that was 44.1kHz/32bit from another unsigned artist. Everything else was 44.1/24.

Link to comment
Share on other sites

  • Members
Quote Originally Posted by Bookumdano2 View Post
No disadvantage that I know of. I've been tracking at 48k since about 2004 or so.

Here's something that I can't figure out...

If you're ever planning on providing hd tracks to something resembling a "market", you better track at 96k or 192 .. like now.

I always figured I could port stuff or upsample to 96 on existing tracks but you know what?... there are guys in high fidelity land that know when you do that. They can tell from scopes and screens and things.

How do they know that? That really throws me. If I do a real-time port of a 48k 24bit song into a second daw at 96k... those guys ALWAYS know that the source was 48.

And then they call foul.

How do they do that?????? I'm going to have to find out more from those expert high-fidelity nerds.

Anyway, I'm probably going to ramble on over to 96k pretty soon anyway for tracking.

There's really no reason to avoid it now I guess.

Bought my first cell phone this year so anything's possible.
It is basically a matter of looking for content that shows a 90 dB or less signal to noise ratio and frequency band limiting of approximately 20 kHz -- probable signs of upsampling from 'standard' format. And a certain sign that you have wasted your money, no matter.

There's been a bit off a mini-scandal regarding such content being sold at HD premium prices from some vendors.
Link to comment
Share on other sites

  • Members
Quote Originally Posted by Kendrix View Post
I believe Cubase samples at 32 float when set to do so..
Sample rate is determined by the audio converter's setting, which one can presumably set from the DAW. However, there are only a very small handful of such converters -- mostly targeting the often over-credulous audiophile market. There is no reason to sample at 32 bits, since your analog front end is probably only capable of the equivalent of about 20 bits of dynamic range, maybe less.

However, many DAWs allow one to save into a 32 bit floating point format in order (as others note above) to allow greater flexibility in transferring works-in-progress from one work station/system to another.

[Of course, that's not to say that there's no reason to increase the bit depth of the math one uses for processing during FX and summing processes. As processing complexity and the number of computations goes up, greater mathematical precision will lead to less rounding error.]
Link to comment
Share on other sites

  • Members
Quote Originally Posted by chris carter View Post
FWIW of the past say 100 or so records I've mixed, unsigned, indie or major label, almost ALL of them have been 44.1kHz/24bit. I got one project that was 48kHz from an unsigned artist and one project that was 44.1kHz/32bit from another unsigned artist. Everything else was 44.1/24.
Electronic Musician did a survey of recording resolutions used by users. The overwhelming majority was 44.1kHz/24-bit, with a few 96kHz folks.

Personally, I don't hear any significant difference between 44.1 and 48kHz. I do hear a difference at 96kHz with certain plug-ins, like amp sims, but I suspect it has more to do with the math behind the plug-in possibly being more precise at higher sample rates. But what do I know? I plug my guitar in, gain-stage properly, and click "record." idn_smilie.gif
Link to comment
Share on other sites

  • CMS Author
Quote Originally Posted by Kendrix View Post
I believe Cubase samples at 32 float when set to do so..
Cubase doesn't sample at all. That's done in the A/D converter, probably an off-the-shelf chip in your interface. It puts out a 24-bit word, which is why it's called 24-bit sampling. Actually anything beyond about the 21st or 22nd bit is just noise, which is why it's sometimes called "24 marketing bits."

If you tell Cubase, or Sonar, or even, I think, Audacity, that the sample is greater than 16 bits long, it writes the data as a 32-bit word for its own convenience. So you have 22 or fewer bits of actual useful data, a few bits of noise, and 8 bits of zeros in your 32-bit chunk. Those zeros will get used as soon as you do any manipulating of the audio so they won't go to waste.
Link to comment
Share on other sites

  • Members

 

Quote Originally Posted by blue2blue

View Post

It is basically a matter of looking for content that shows a 90 dB or less signal to noise ratio and frequency band limiting of approximately 20 kHz -- probable signs of upsampling from 'standard' format. And a certain sign that you have wasted your money, no matter.

 

That's what I was thinking. But I've never put a high resolution recording on a frequency analyzer to see if there is anything recorded above 20kHz using typical mics. If there's nothing there anyway, it would be hard to tell if it had been band limited or not.
Link to comment
Share on other sites

  • Members

I record at 44 and upload to Soundclick. It's very apparent to me that at a certain ranking, I hit a wall, and that wall was recording quality.... and I go direct to DAW...the guys above me in rankings, all things equal, just sounded better.

The only thing I could do was get a better DAW and maybe better software and maybe a better PC, which is all about computing that sampling and data at a higher resolution, more accurately, with less issues.

Can you hear the 44 vs 48 difference? Throw up your music on Soundclick, see how high it goes...if you get to say the top 20 and find everyone just sounds better then you at the higher rankings...you tell me what else you can do other then to get the gear that makes you sound even better.

Link to comment
Share on other sites

  • Members
Quote Originally Posted by Sillypeoples View Post
.the guys above me in rankings, all things equal, just sounded better.

The only thing I could do was get a better DAW and maybe better software and maybe a better PC, which is all about computing that sampling and data at a higher resolution, more accurately, with less issues.
There are dozens of factors that will influence the audio quality of your recordings more than the DAW or PC that you use. They include:
  • the quality of the sound sources
  • microphone choice
  • mic placement
  • recording room acoustics
  • preamp quality
  • the quality of any devices between the preamp and the audio interface
  • the quality of the audio interface to your computer
  • the recording levels you recorded with
  • the quality of the plug-ins you use
  • your ability to do a good mix
Link to comment
Share on other sites

  • Members

 

Quote Originally Posted by blue2blue

View Post

It is basically a matter of looking for content that shows a 90 dB or less signal to noise ratio and frequency band limiting of approximately 20 kHz -- probable signs of upsampling from 'standard' format. And a certain sign that you have wasted your money, no matter.

 

Seems more a conversion detection process, doesn
Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.


×
×
  • Create New...