I haven't touched on this for a while...
There are two ways to "bit reduce" audio:
1 - Truncate. This is where the least significant bits (they turn out to be VERY significant...) are just not transferred. In the case of going from 24>16 bits, the lowest 8 bits would not get transferred. Only the 16 bit "above" it per se.
2 - Dithering. Dithering adds noise to the audio at a certain dB. For 16 bit dithering, it is around -93dB. Technically speaking, 16-bit audio is capable of 96dB of dynamic range. So, by applying dithering, you technically lose sound-to-noise-ratio (this can be LESS than 96dB on bad A/D/A converters!!!).
Now, you may be a wonderin' why you would want to add noise rather than just lose those "lower" bits. That is the hard part to explain. I will take a shot at it as I have not written a long essay on here for a couple months, and I am due!
Truncating is bad! Why? Simply because those "least significant" bits hold valuable information in the audio the make the lower volume stuff contained within it sound more "real". You ears are capable of hearing a LOT more dynamic range than 96dB. Even though most equipment may have sound to noise ratio's that are less than 96dB, that does not mean that it is impossible to record a dynamic range that exceeds it. It just means that the lowest volume stuff will have a little bit of noise added to it. Most of us are really lucky to get real world 80dB of "clean" audio, where you have 80dB of s/n/r (sound to noise ratio...) So, we can hear more that 80dB of dynamic range, just that anything that is more than -80dB will have a little added noise to it from the electronics. No big deal really except if you want to record an orchestra. Even then, you can still achieve outstanding "depth" in the recording with that kind of s/n/r.
So why is this lower bit info important to audio? This is hard to explain, and I won't attempt to give "technical" explanations, just a wrap up on what was discovered by the "experts" the studies on it.
The lower bit info is important because it is what gives the recorded track a sense of "space" and "depth". If you are recording at 24 bit, you are "real world" on most converters getting around 104dB of dynamic range (I know, the specs on the converters claim much better than that...but if you have been around for a while, you KNOW that specs mostly lie in favor of making the equipment look better on paper then it sounds in real life...) That is the way of marketing these days, and that is not going to change!). So, you have about an extra 8dB of dynamic range that is possible with 24bit over 16 bit. Cool! This means that the quietest to loudest sound recorded is almost DOUBLED with 24 bit (approx. every 10dB increase/decrease in audio = doubling/halving of the perceived volume...) This improved dynamic range mean one can accurately capture reverberation in the room, which helps give the recorded sound a more "real" flavor. Cool!
Another thing about digital audio. This is important!
Bit 1 only has 2 options: On and Off. What this means to you while recording is that stuff that is recorded at the lowest levels of the dynamic range has few options for being recorded. If for instance, your recorded part has sonic information that hovers around the bottom of the dynamic range of the converter, if it turns down even a little bit, that last bit of information may or may not activate. The audio may have slipped BELOW. This can be a disaster during note's that fade out!!! The more dynamic range we have, then better, because our ear is VERY sensitive to low volume sonic information! More dynamic range available = truer representation of the original sound! So, if the fade out is hovering around -104dB and say drops down to -105, well, that sonic information is lost! Oops......
So, truncating is bad! You lose low level information in audio! Your audio suffers from sonic "space" because when you originally recorded it with a higher dynamic range available, you made "decisions" about the "quality" of the audio based upon a superior dynamic range. The whole sense of space in the sound will be mostly lost when you remove those "least significant" bits. To your ears though, they become indeed "significant".
Dithering! Can't live with it, can't live without it! "Quote by Bob Katz. The internet GOD of digital audio......"
Some smart guy thought out all that stuff about losing lower bit information and had an idea that there may be a better way! Thank god!
Dithering adds a CONSTANT "shaped noise" (noise at certain frequencies. Limited band.) at a certain -dB. In the case of 16 bit, it is around -93 dB. Why does it do that? This is clever!
We know that as the lower bits are only used, like in the case of a fade out, that there are less bits to capture small changes in the volume. When you are recording to digital, the audio is actually "dithered" while recording because the voltage at the A/D converter may fall between two bits. So, the voltage is averaged up or down to fit a certain bit. If you are recording "hot" to the A/D converter, this is far less of a problem because the "upper" more "significant" bit have finer division to represent the audio. So, when the voltage drops, there are less options for accurately capturing the sound. Thus, when it is dithered while recording, lower level stuff will have "quantinization errors" which tends to sound like distortion because the representation of the input voltage is actually getting shifted up or down so that the audio can be stored to a fix bit. This is usually only noticeable at on the last bit or two or three. Above that, you cannot hear the quantiniztion errors as distortion.
So, if a constant noise is applied at a certain -dB, Those bits will always be "on". This is important. This means that even though the input voltage falls below what the converter can deal with, the last bit is always on, so any audio that is below the dynamic range of the last bit can still be stored to the last bit. The incoming voltage and noise are "mixed" together. So, the noise is actually keeping that last bit on all the time and blending the music with it, so very quiet stuff can be recorded to the last bit, and you can still usually hear some volume difference! VERY COOL!!! We are cheating here! We are improving the DYNAMIC RANGE of the converter so we can have music that exceeds it's specs! Wow!!!
Oops! There is a trade off. Noise!!!
Yup, we ADD noise with dithering. But, this noise is "shaped" noise, meaning it is not "across the whole frequency spectrum. Shaped noise in dithering is in fact usually out of the most sensitive region of our hearing, 2KHz-5KHz. This does not mean we cannot HEAR the noise, it just means that we are less apt to hear it as a noise that conflicts with musical content. At very low volumes, like around -93dB, we would have to have out speakers pretty darn loud to hear say a 10KHz noise. But, we could still hear the 2KHz-5KHz stuff very well at a much lower speaker volume. Refer and study the Fletcher/Munson Relative Loudness Curves to see what I mean.
So, dithering adds that noise, and we may even hear it, BUT, the noise is not distracting. So, that "nice" noise in this case is keeping our "lowest bit" activated to "on" all the time, thus, incoming audio can be "blended" with it and stored to that lowest bit, and still retain dynamic range. This gives the illusion of more dynamic range. Cool!
Now, it is important to remember that dynamic range and signal to noise ratio SORT OF work hand in hand. A A/D/A converter can have excellent dynamic range specs, but if the noise it creates is in the most sensitive part of our hearing, 2KHz-5KHz, the effective dynamic range is effected! Signal to noise ratios are ALWAYS LOWER than the dynamic range capability. But, if the noise is in a less sensitive region of our hearing, than the noise will not "mask" incoming audio. Cool eh?
So, when you have 24-bit audio, you have outstanding dynamic range. You can really hear DEEP into the sound! But, if you truncate that dynamic range to 16 bit audio, you lose a lot of that low level information, which tends to decrease the amount of "space" and "depth" to your recording. Oops. But, dithering adds that constant noise at a range of frequencies that are out of the ears sensitive region, so, you can actually retain quite a bit of that dynamic range. Say maybe around 19 bits if the audio was recorded quiet enough (not so much noise from the electronics).
So, when you dither, you add noise, but can potentially gain dynamic range. You add dithering before you truncate the audio, so that the least bit contains more actual sonic information.
Okay, enough of that. On to other things.
Sometimes, your recording may be so noisy that you may be “self dithering” it. Meaning that your combined noise from electronics (preamps, room noise, etc….) are keeping a constant noise going at a certain –dB. Dithering after the fact may not do anything other than add MORE noise. So, in some cases, and mainly with very low end gear that most home recording guys use, dithering MAY NOT be necessary because you do not have an effective dynamic range to your recorded tracks due to sound to noise ratios being pretty bad. If the noise you are introducing to the audio is in a sensitive part of your hearing, it will mask audio content there anyway, so you effectively have less dynamic range available. Get it? So, while S/N/R and Dynamic Range are two different things, one effects the other.
High-end audio gear tends to have much better s/n/r than cheaper gear, thus, is capable of more dynamic range. This is the reason why high dollar equipment tends to provide sound that has more “depth” and “space” to it, because it is not creating noise in your sensitive hearing region that masks musical content at very low volumes. You can hear reverberation in the room much better, and this helps keep the sound more real sounding. Remember, you ear is VERY sensitive, and has a very wide dynamic range available to it (around 19-20 bits).
The only way to actually determine whether you need to dither your 24 bit audio or not is to try it both ways. Truncate a version of the mix to 16 bit, and dither another version of it. Listen to the quietest passages at very high volumes and listen. Does the truncated version sound like it has less depth and separation at low volumes compared to the dithered version? If so, you will want to dither. If not, you probably have a noise floor that is higher than what dithering would do, this, dithering may not be needed.
Enough of that. Use your ears. Decide based on that what is best for your audio.
About plug-ins that dither.
The Waves L1 Ultramaximizer has a pretty good sounding dithering scheme to it.
The Wavelab dithering scheme isn’t too bad either. I like the sound of the Waves L1 better though. Remember, there may be even better ones available, but usually, only in TDM format (for ProTools) and in stand alone D/A converters.
Hope this helps a bit.
Ed