Wednesday, April 18, 2012

24 Bits, 96 kHz: What Does It All Mean?

Ok, so the status quo for digital recording at the moment is 24/96.  That would be 24 bits at 96 kilohertz.  This is a fair step up from when I got started.  CDs are 16/44.1 and most of the stuff around at the time was as well.  Why make the jump and fill up your hard drives faster?  Two reasons.

Let's look at bit depth first.  If you look at each digital word that represents a sample it's really a pretty simple animal.  If you had a one bit system, you get a 0 for everything quieter than half the level the system is capable of taking in, and a 1 for everything over that.  There's no in between.  If you keep adding on bits you get smaller increments that are capable of being recorded.  But why only 16?  If you think of all the sound that takes place in any recording, a ruler with only 16 notches on it seems like a pretty puny tool for the job of faithfully recording it.  But if you take your measurements really fast things get better.

Without getting into a long scientific discussion let's just cover all this stuff about "headroom" being better in a 24 bit system and move on.  If you're working at 16 bits and you record something with a lot of dynamic range, you have to hold the volume down so that the peaks don't run out of room.  You wind up with a lot of quiet sounds in there. Later on if you're looking to normalize or compress that signal, there are only a few bits describing those quieter sounds.  If you record at 24 bits then you don't need to record as hot because there are more bits describing the quieter signals.  You can raise them up in volume and maintain better quality.  The only thing you need to watch out for is converting down to lower bitrates.  Some poorly implemented converters accomplish the job by just lopping off the least significant bits, good ones do a lot of math to make sure everything translates in the best possible fashion.

Now let's get on to sample rate.  The Nyquist rate is the speed at which you need to sample to obtain frequencies that are half of that rate.  So a 44.1 kHz system can capture audio frequencies up to 22.05 kHz which is above the hearing range of humans, so we're all good right?  Not so fast.  There are frequencies emitted by almost any source that are above our hearing range.  What happens if you take a sine wave at 25 kHz and another one at 26 kHz and play them back?  They're both above our hearing range, but you would actually hear a beat frequency where the two interact at exactly 1 kHz which is right in the middle of our hearing range.

There are mics now that are capable of capturing frequencies in excess of 100 kHz.  Pair them up with a 96 kHz recording system and you're capturing just a ton of data that we can't hear, but some of the interaction between those frequencies up there will translate into stuff that we can hear.

So it's not hard to see why it's worth making the jump up to a 24/96 system or even higher.  This is one of those areas where if good enough isn't good enough, you can take some extra effort that might seem superfluous but will actually improve your work in subtle ways.  Why?  Do low bitrate MP3s sound OK?  That depends, do you just want to hear that catchy song from the radio and enjoy the hook on your earbuds?   Or will you be trying to reproduce that same song in a venue with pristine line arrays for a discerning audience?  I'd be looking for higher bit and sample rates myself in the second situation.

Hopefully that sheds a little light into the murky depths of the digital audio realm for my fellow Brethren of the Knob and Fader.  Feel free to hit the comments section or contact us through Facebook or Twitter to further the subject.


  1. You had mentioned 16bit technology and compared to a ruler with 16 notches. I'm not sure if you meant to imply that there were only 16 discrete values for volume information, but that's how I read it. As I understand it, this is not an accurate description. I'm not a computer guy, I'm a sound guy, but a sixteen bit system would mean four bytes. So the first value could be described as 0000 0000 0000 0000, a second value as 0000 0000 0000 0001, a third one as 0000 0000 0000 0010, and on and on and on. So the proverbial ruler has not 16 notches, but rather 65,536 (incidentally 24 bit= 166,777,216 and 32 bit= 4,294,967,296) notches. Maybe this is wrong, but I don't think so because if it were true that there were only 16 values, it would be impossible to use software (plugins) to process dynamic effects. Gates and compressors would so inaccurate that they'd be unusable. This is obviously not the case. This is not even broaching the topic of getting a healthy signal to error ratio, which is somewhat similar in practice to getting a good signal to noise ration in the analog world as you touched on in the post. You said, there are many, many samples taken each second and that is, of course, true, but that would have little baring on the accuracy of volume reproduction as sample rate is a different parameter altogether and determines the accuracy of frequency reproduction, not volume.

    Ultimately, yeah, like you said, there aren't really too many reasons to not record at the highest bit depth as is available to you, which was the point of the post. I just wanted to throw this out and get your opinion.

  2. Yeah, it's been over a decade since I read Principles of Digital Audio so my bit math is a little rusty. Every time you add a bit to the word length you double the resolution. A one bit word is either 1 or 0 and cuts the signal in to everything above 50% and everything below. A two bit word can be 00, 01, 10, or 11 giving you four steps. Keep multiplying by two and you get to 65,536 like you said. Still not a lot of depth considering how sensitive the ears are. Looking at color depth on a monitor you can see the difference, you can hear it as well on good equipment.

    I think you slipped a decimal somewhere because I got 16.7 million for a resolution at 24 bit, but my math agrees with yours at 32 bit. But back to the equipment. In a 16/44.1 world where the CD is king things are good enough, but you can sure tell the difference when you listen to higher depth/rate masters. But then you go in the opposite direction and how do things work out for low bitrate MP3s on cheap earbuds?

    What I hear from bigger dogs than me is that 24/96 is the way to go right now. Working higher than that can have advantages in certain cases, but for the penalty in storage and processing power it's not worth it a lot of the time. Where the technology is right now, 16/44.1 to 24/96 is like going from decent monitors to expensive ones. Audible but not earth shattering. But going beyond that, the difference is arguable.

    1. That brings up a WHOLE 'NOTHER can of worms; an argument that I frequently find myself on the losing side of: Digital audio vs. analog, or more relevantly, compressed digital audio vs. full rez. At what point and/or under what circumstances is it "good enough".

    2. When the client stops hearing the improvements you're making?


You're the Scotty to our Kirk