Whats the difference between recording at 16, 24 or 32 bit?

[mrmeggs]

Hi ive been recording at 32 bit 44000hz (i think) is this ok? Does this make any difference to the sound quality? What would happen if i changed this?
Cheers, HMMMM?

 

[RhythmRmixd]

It's impossible to record anything at 32 bit, you can mix and process in 32 bits, but once you record anything to an audio file you have to reduce bit depth to 24 or 16. Either you need to check your documentation for your equipment, or I need to get out of the house more and find whatever converters your using. I could use that kind of headroom.

 

[DonF]

The difference is in the accuracy of the digital representation of the analog waveform. Higher bit depth gives you greater resolution in the vertical dimension, which gives you greater dynamic range, and more options later on for manipulating the dynamics of your recording (compression, etc.). Practical application: If you're recording with a 24-bit depth, you don't have to record as hot as you would with a 16-bit depth, so you don't have to risk "overs", which means fewer retakes.

If you have a choice, go with 24- or 32- bit depth rather than 16. If your finished product will be a CD, you will eventually have to convert it to 16-bit. This step should happen as late in the game as possible.

Don

 

[mshilarious]

It's impossible to record anything at 32 bit, you can mix and process in 32 bits, but once you record anything to an audio file you have to reduce bit depth to 24 or 16. Either you need to check your documentation for your equipment, or I need to get out of the house more and find whatever converters your using. I could use that kind of headroom.

If you need 32 bit headroom (192 dB), you should be completely deaf by now

You can have 32 bit audio files, of course because converters are only 24 bit, there would only be 24 significant bits, but there is nothing stopping you from saving a 32 bit file.

 

[DonF]

It's impossible to record anything at 32 bit, ...

Lots of modern DAW's store audio data in a 32-bit floating-point format. You are correct that I/O must be done in 16- or 24-bit (or whatever the converters support), but files can be saved in a 32-bit format, and there are advantages to doing this.

Don

 

[reshp1]

You will have 2^32=4,294,967,296 steps of resolution between line level 0 and silence with 32 bit

You will have 2^24=16,777,216 steps of resolution between line level 0 and silence with 24 bit

You will have 2^16=65,536 steps of resolution beween line level 0 and silence with 16 bit

16 bit to 24 bit is like normal TV to HDTV

24 bit to 32 bit is like HDTV to real life.

 

[gusfmm]

but files can be saved in a 32-bit format, and there are advantages to doing this.
Don

Such as??? Please be explicit, I feel like watching a 256 color picture in my 19in 1280 x 1024 Sony SDM-S94 flat LCD at 16.7M colors... same crap no matter all the greens I paid for the bloody LCD.
Sampling is still going to be 24bits, regardless of whatever the file format you save to happens to be.

 

[mrmeggs]

Thanks guys I had no idea

Thanks for the info. I am using Nuendo and M-audio 10/10s and down the bottom it says 32 bit so I suppose it makes sense that they are converted once recorded. So I suppose that recording at this setting is fine? It can be changed later? What about recording at double the hz what does this do. I suppose this would take up much more space but would it increase the sound quality a great deal?

 

[DonF]

Such as??? Please be explicit, I feel like watching a 256 color picture in my 19in 1280 x 1024 Sony SDM-S94 flat LCD at 16.7M colors... same crap no matter all the greens I paid for the bloody LCD.

Until you take that 256 color picture into a digital video editor and try to fade it to black. Compare your cruddy 256-color fade to the one the DV editor creates after you've converted to a better format. It will look better with the conversion. Trust me on this. Similarly, if all you want is to copy your audio from a source to a destination, there is no sense in using more bits. But if you want to do any manipulation of the signal, do the conversion. Disk space is cheap.

(edit) I forgot to mention that most computers these days are better at handling 32-bit chunks of data than they are at handling 24-bit chunks of data, which means that you can potentially get better performance if the application uses 32-bit data internally, regardless of the format used for I/O.

Don

 

[dgatwood]

Bear in mind that you're talking about 32-bit floating point versus 24-bit integer. Floating point numbers are stored as the equivalent of scientific notation, roughly (except in base two, but that's not important for the purpose of this explanation).

For example, 104 can also be expressed as 1.04 * 10^2. The 2 is referred to as the exponent, the 1.04 is the mantissa (roughly).

The result of this is that floating point numbers can handle a wider scale than fixed point numbers, at the cost of reduced precision within that range (since 1.04 may not be precisely representable in binary... not sure on that particular value...).

For convenience (and for computational reasons), most effects processing is done using floating point numbers. Among other things, this means that wide ranges of loudness can be represented without the need to reduce the level at any point in the effects chain (except at the end when you end up converting it into an integer to export to a file and/or to actually play it on your computer's speakers).

An IEEE 32-bit float turns out to be a 24-bit mantissa, so it should be (roughly) functionally equivalent to doing 24-bit audio, just with orders of magnitude more headroom in the effects chain.

Now I'm not entirely sure why some audio programs choose to store data on disk as 32-bit float. Could be laziness not wanting to do it at playback time prior to mixing multiple tracks, or it could be that they assume you might do multiple effects passes on a single file and resave it each time, and this essentially lets you avoid worrying about clipping and also avoids loss of precision from multiple integer<-->float conversions. Dunno.

Either way, take this with a grain of salt, as I'm half asleep.

WRT recording at a higher sampling rate, the nature of analog to digital conversion involves rolling off the audio sharply at the Nyquist point (half the sample rate). This avoids aliasing---false, low frequency signals equal to... IIRC the frequency minus the Nyquist point. For an example of aliasing, watch a car tire appear to spin backwards on TV or in a movie. Same issue.

Anyway, recording at a higher rate shifts the Nyquist point higher, and the filter cutoff along with it. Since filters aren't perfect (particularly cheap analog filters... ) and may start to roll off high frequencies significantly below the Nyquist point (or else you end up with a noise "bump" right below the Nyquist point), recording with a higher Nyquist point should mean there's less chance of that filter ending up rolling off some of your high frequencies along with the undesirable 22kHz+ noise.

That said, I'm not sure exactly what effect downsampling to a more sane rate has, nor how aliasing is avoided in that process. I'm guesssing FFTs and stuff do a better job than analog filters, but somebody with a degree in EE could probably do a better job explaining that....