|
16 bit vs. 24 bit AUDIBLE DISCUSSION (NO TECH SPECS) (pg. 3)
|
View this Thread in Original format
| coroknight |
When you are mixing your tracks their are all sorts of things that change the dynamic range of each channel. Even setting the volume lower than max reduces the dynamic range. Also, effects change the dynamic range, some effects will cause the volume of a signal to drop and then compensate by boosting the signal before outputting it.
This also lowers the dynamic range because, lowering the signal rounds off certain values.
Luckily, DAWs generally operate in 64-bit so that when you lower the dynamic range you still have a MASSIVE amount of range left.
However, 16-bit is ok for the finished product since we won't be manipulating the signal after that point.
Also, by finished product I mean after mastering. It's still a good idea to have a really high quality render pre-master. |
|
|
| DigiNut |
| quote: | Originally posted by DjStephenWiley
But beyond classical, I can still hear a difference in other things. That difference is clarity, and it's quite noticable with my AKG 701's which are known for having some of the best hi-end hifi monitoring available. Lately, 24 bit or 16 bit has been a deciding factor in buying sample packs. |
"Clarity" is not a quantitative term.
It's usually a good idea to record at 24 bits because you have a lower noise floor, and input devices like mics tend to give out a small signal. But as for bouncing to 24 bits and then encoding as a 16-bit MP3, you're going to have to do better than that for an explanation/evidence; most likely what you're hearing is not clarity, it's distortion caused by dithering (or worse, flat-out truncation), that you've conditioned yourself to subjectively prefer.
Nothing wrong with that as long as you recognize it. In a similar vein, a lot of people prefer the sound of a classic tube amp to a digital amp because the tube amp produces 2nd-order harmonic distortion and colours the sound.
Your sequencer internally is already mixing at 32 or 64 bits (floating point), so there is absolutely no way you can logically claim to get more "clarity" from bouncing to 24-bit as an intermediate on the way down to 16 bits than you would from simply dithering directly to 16 bits. It's actually no different from claiming that you can make a better vinyl by recording onto a cassette tape first - although I would imagine that more people realize how ridiculous that sounds because it's not as technical as "24 vs 16".
When you downconvert from 64/32 to 24 bits, you lose some fidelity. When you downconvert again from 24 to 16 bits, you lose more fidelity. The cumulative loss from two conversions is, almost as a matter of definition, worse than the loss from a single conversion from the same source to the same final destination. If it happens to produce a pleasing effect for you, that's a coincidence. |
|
|
| coroknight |
I believe a college professor did a study where every semester he had students listen to an MP3 song and then the same song in WAV format. He then recorded how many people preferred each version.
The conclusion was that the distortion of the MP3 was more familiar than the sparkling shine of the WAV and generally people prefered the MP3 to the WAV. |
|
|
| DjStephenWiley |
| quote: | Originally posted by DigiNut
"Clarity" is not a quantitative term.
It's usually a good idea to record at 24 bits because you have a lower noise floor, and input devices like mics tend to give out a small signal. But as for bouncing to 24 bits and then encoding as a 16-bit MP3, you're going to have to do better than that for an explanation/evidence; most likely what you're hearing is not clarity, it's distortion caused by dithering (or worse, flat-out truncation), that you've conditioned yourself to subjectively prefer.
Nothing wrong with that as long as you recognize it. In a similar vein, a lot of people prefer the sound of a classic tube amp to a digital amp because the tube amp produces 2nd-order harmonic distortion and colours the sound.
Your sequencer internally is already mixing at 32 or 64 bits (floating point), so there is absolutely no way you can logically claim to get more "clarity" from bouncing to 24-bit as an intermediate on the way down to 16 bits than you would from simply dithering directly to 16 bits. It's actually no different from claiming that you can make a better vinyl by recording onto a cassette tape first - although I would imagine that more people realize how ridiculous that sounds because it's not as technical as "24 vs 16".
When you downconvert from 64/32 to 24 bits, you lose some fidelity. When you downconvert again from 24 to 16 bits, you lose more fidelity. The cumulative loss from two conversions is, almost as a matter of definition, worse than the loss from a single conversion from the same source to the same final destination. If it happens to produce a pleasing effect for you, that's a coincidence. |
Post some samples then. High frequency percussion with a large stereo field. Take an identical recording at 24bits and convert it over to 16, then convert both to 320 mp3 and post them here. You may want to do a couple of these. |
|
|
| MrJiveBoJingles |
| Whether a high bit rate will be best for a sample has absolutely nothing to do with its high frequency content. High frequency content will be affected by sample rate, not bit rate. Do you know what the function of bit depth is in digital audio? |
|
|
| coroknight |
| Hi frequency percussion is probably not a good example since percussion is very noise'y. |
|
|
| Beatflux |
| quote: | Originally posted by coroknight
I believe a college professor did a study where every semester he had students listen to an MP3 song and then the same song in WAV format. He then recorded how many people preferred each version.
The conclusion was that the distortion of the MP3 was more familiar than the sparkling shine of the WAV and generally people prefered the MP3 to the WAV. |
I have had a similar experience where I preview something on Beatport, I download the 320kpbs file and I'm disappointed because the low bit rate hid ugly elements or meshed them together so I couldn't hear them. |
|
|
| Storyteller |
| quote: | Originally posted by DjStephenWiley
Post some samples then. High frequency percussion with a large stereo field. Take an identical recording at 24bits and convert it over to 16, then convert both to 320 mp3 and post them here. You may want to do a couple of these. |
I laughed when doing this:
First the 24bit part, then 16bit, then an inverted 16bit against the 24bit (which should result in the audible difference). No dithering applied. However it's all been done within the sequencer to have 32 bit processing.
http://www.storytellermusic.nl/16bitvs24.mp3 |
|
|
| DigiNut |
| quote: | Originally posted by DjStephenWiley
Post some samples then. High frequency percussion with a large stereo field. Take an identical recording at 24bits and convert it over to 16, then convert both to 320 mp3 and post them here. You may want to do a couple of these. |
That is not the right process. You want to be comparing the original production in 32/64-bit float, properly dithered directly to 16-bit using an industry standard like UV22, and compare that to the same production dithered to 24 bits using the same standard and downconverted again to 16 bits using whatever process you choose.
What you're effectively saying is that you can tell the difference between a 24-bit recording dithered to 16 bits and the same recording truncated to 16 bits by some MP3 encoder. That is obviously true or at least believable, but it's not particularly interesting because you've already introduced the main lossy step (downconversion to 24 bits).
So how about this instead: You find some projects you think are appropriate for the test that are in the raw sequencer format (we'd need at least 5), and export each as a 32-bit floating point .wav, which will be a straight bounce with no downconversion. I'll take that and make two 16-bit versions encoded as 320 kbps MP3, the first in one stage, the second in two stages (dither down to 24-bit first). Then we will do an ABX test, to see if the two-stage version is actually identifiable at all.
Assuming we actually get past the ABX, which I doubt, I'll take each track, including the original, toss it into Wavelab, and post the statistics on each, and we can see which one is actually closer in fidelity to the original.
I always refuse to use my own tracks for these tests because it opens the door for excuses about the source material. You pick the tracks, and they have to be posted as 32-bit or 64-bit float PCM (.wav) files - if they're already 24-bit then half the test is out the window. |
|
|
| DjStephenWiley |
Well I'm really not concerned about anything above 24bit. I know I brought up 32bit, but 24 bit is my main focus and concern (hence the topic title)
I'm not going to flip out and you say you tainted the source and purposefully tried to fool me using your knowledge to prove you're right and i'm wrong. I have plenty of respect for you and I'm sure you're as eager as I am to see if I can differentiate between the two. Just get some 24 bit samples (get a variety, low mid and high freq's) and encode them to mp3 from 24 and 16
To the poster asking regarding high frequencies and bit rate, of course I know the difference. Jeez. I can make music but it takes a lot more time and effort for me than the average bear. |
|
|
| palm |
| quote: | Originally posted by coroknight
He then recorded how many people preferred each version.
|
did he use 24 bit or 16 bit? :p |
|
|
| DigiNut |
| quote: | Originally posted by DjStephenWiley
Well I'm really not concerned about anything above 24bit. I know I brought up 32bit, but 24 bit is my main focus and concern (hence the topic title) |
With all due respect, I think that you're missing the point.
The source is ALREADY in 32-bit. If your source material is a 24-bit sample, then the most significant downconversion (from floating-point to fixed-point) has already happened, and your comparison is simply one of dithering vs. truncation (which is what your MP3 encoder will do). These will obviously sound very slightly different, but the truncation (the one you identify as being clearer) is actually the one with worse fidelity. Think of anti-aliased text or graphics that you usually see on screens today vs. the old-school machines where you could always see the jagged edges - it's the same concept.
It's just not a valid test to start from 24-bit unless the entire sample is recorded from a live source into a 24-bit ADC. You need to start with the same bitrate that's native to the production environment, and that's either 32 or 64 bit floating point depending on your PC.
If it's what you want then I'll take a couple of old tracks and upload the samples that I specified in the earlier post - one dithered directly to 16 bits and another dithered to 24 bits and then dithered again to 16 bits. No truncation, i.e. trying to encode a 24-bit sample to MP3 - we already know why that sounds different. |
|
|
|
|
