Bit depth

David G. Smith

David G. Smith

Steak Knife Member
Ok, I am a camera guy and, have a question about audio. Most audio for film/video is recorded at the 48.00 Khz sample rate, correct? 16 bit is the standard bit depth for most video and film applications, correct? Now, does the bit depth, 16 vs 24 make any real world noticieble difference? Would there be any reason to chose to record film or video audio (double system sound) at the higher bit depth? Is bit depth anagalous to resoulution, when talking about images? Can most NLEs handle audio at different bit depths?

Thanks.
 
Bit-depth matters.

The basics:

Digital audio is not a continuous, analogous stream (like analog is). The device takes a sample of the audio x times per second, and upon playback the device has to fill in (a best guess) between the samples... a sort of connect-the-dots game. At 32kHz, the device samples the audio 32,000 per second. At 44.1 kHz, it's 44,100 times per second. At 48kHz, it's 48,000 times per second. The higher the sample rate, the more "dots" there are and the less that has to be connected between them. That means higher fidelity.

Alongside sample rate is bit-depth. Each sample is made of a word, and that word is made of bits (1s and 0s). So at 16-bit, each word is 16 characters long. At 24-bit, it's 24 characters long. The longer the word length, the more detailed the sample. At 16-bit, if your audio drops below a certain level, you may only be using the first 8 bits of the word. At 24-bit, you'll use 12 bits for the same signal. This really makes a difference in post*, as you can process that signal with fewer artifacts if you have more data with which to work.

Your NLE may or may not like 24-bit, but you can always down-rez to 16-bit without losing quality, and replace the 16-bit with the original 24-bit sources when you go to audio post.

*Case in point: I recently did post on a show for one of our local affiliates. They had recorded an intervew without a sound person, and had done a terrible job of it. The levels were barely bumping the meters, and it was 16-bit audio. I tried to bring the levels up (which amplified a nasty amount of hiss) and to use some noise reduction to reduce the hiss. Because the audio only utilized a few of the bits, the more I processed it the more "robotic" the sound became. Yeah, it is analogous to camera resolution... if you have a low-rez image and you begin to process it in PhotoShop by enlarging it and then adding effects, you'll start to see the pixels more and more. In the case of audio, you start to hear the "pixels."
 
Last edited:
C2V said:
Bit-depth matters.

The basics:

Digital audio is not a continuous, analogous stream (like analog is). The device takes a sample of the audio x times per second, and upon playback the device has to fill in (a best guess) between the samples... a sort of connect-the-dots game. At 32kHz, the device samples the audio 32,000 per second. At 44.1 kHz, it's 44,100 times per second. At 48kHz, it's 48,000 times per second. The higher the sample rate, the more "dots" there are and the less that has to be connected between them. That means higher fidelity.

Alongside sample rate is bit-depth. Each sample is made of a word, and that word is made of bits (1s and 0s). So at 16-bit, each word is 16 characters long. At 24-bit, it's 24 characters long. The longer the word length, the more detailed the sample. At 16-bit, if your audio drops below a certain level, you may only be using the first 8 bits of the word. At 24-bit, you'll use 12 bits for the same signal. This really makes a difference in post*, as you can process that signal with fewer artifacts if you have more data with which to work.

Your NLE may or may not like 24-bit, but you can always down-rez to 16-bit without losing quality, and replace the 16-bit with the original 24-bit sources when you go to audio post.

*Case in point: I recently did post on a show for one of our local affiliates. They had recorded an intervew without a sound person, and had done a terrible job of it. The levels were barely bumping the meters, and it was 16-bit audio. I tried to bring the levels up (which amplified a nasty amount of hiss) and to use some noise reduction to reduce the hiss. Because the audio only utilized a few of the bits, the more I processed it the more "robotic" the sound became. Yeah, it is analogous to camera resolution... if you have a low-rez image and you begin to process it in PhotoShop by enlarging it and then adding effects, you'll start to see the pixels more and more. In the case of audio, you start to hear the "pixels."
Click to expand...


For people who like dB's, 16 bits gives a 96 dB dynamic range, while 24 bits gives 144 dB range.

The problem with these theoretical numbers is one has to also consider the noise of the system as well, and whether the increased bit depth actually gives the desired results.

To be sure while 'processing' the data, choosing the deepest bit depth will allow for computations to not overflow the representation, aka, clip.
 
Thanks. So the file size for a 24 bit recording, at 48k is 34% larger than a 16 bit recording (or however the math works out), correct?
 
David G. Smith said:
Thanks. So the file size for a 24 bit recording, at 48k is 34% larger than a 16 bit recording (or however the math works out), correct?
Click to expand...

I don't think I've worried about audio recording size these days... but a quick look on WAV format suggests that a 24 bit recoring uses 3 bytes for the 24 bit representation (that is doesn't at an extra byte to round to 4 bytes...), so that would be about 50% more data...

This would be for an uncompressed data stream...
 
bit depth is like latitude on a camera.
it does not give you more resolution (that would be more like frequency sampling) but allows to differentiate better between low and high levels.
At the end it is still participating to give a better sound.
currently, we can get sampling at 48Khz, that is a lot regarding what an adult hear can get (usually up to 15 Khz if you are in good health).
And ear is sensitive to low frequencies (between 1Khz and 8Khz), so going up to 15 is not adding a lot more of information.
In the other hand, you get bit depth.
At 16 bit, it means you get 65536 levels. That could seems a lot, but unfortunately, sound is one of these non-linear phenomenon, where things do not simply adds up. So to get a sound twice stronger, you need to multiply by 10. You see the problem, many of these 65536 levels are too close to be useful.
 
Interesting discussion:
I'd just like to add that working in a 24 bit session (as noted earlier) provides more headroom. Which can come in handy as you start to add FX, Foley, Music and more elements of varying levels to your production.

Also, many mixers feel that a 24 bit session helps with the "clarity" of certain audio plug-in effects like reverbs and in some cases (not all) Eq.
 
Thanks all. That makes a lot of sense, and, with no down side, of any consequence, It would be crazy not to do my recording in 24 bit. I appreciate the responses.
 
You must log in or register to reply here.
Back
Top