16 bit or 24 bit?

Otis Grapsas said:
Volume changes are single multiplications with a float parameter. You can probably have hundreds of these in a row without an audible change in quality. The software should be doing them in real time because they cost nothing (cpu wise). They are usually implemented at the last stage of processing in real time and are always active regardless of volume setting. They are destructively processed only on mixdown....
Click to expand...

As I said, depends on how you define it. The stream IS destructively changed at each level point. There is no effective difference between altering the stream an saving the altered file and then processing it again with the next level change. Also YOU may not think it's a big deal but jump over to some professional audio boards and the issue of mixing and math is HUGE.

But pretty much everyone has answered the posters question with the same answer, 16 is fine. We differ on a lot of particulars that he/she are probably not all that interested in.

Happy Holidays
Cheers
SK
 
tyreeford said:
. 16-bit isn't reall 16-bit either, It's usually more like 12-14 bit.
Click to expand...


Most (all?) 24 bit recorders use always 24 bit ADC and dither down to 16 bits when recording at 16 bits. That means that it as good 16 bits as possible (if 24 bit ADC is anything better than 16 bits, that is).

That gives best possible 16 bit resolution at 15.5 bits + 0.5 bit dither. This is better than clean undithered 16 bits.

Example: Canon XH-A1 gave me more than 90 dB dynamic range when fed from SD302. That is more than 15 bits clean signal and the test procedure was not best possible (quieter room would have helped). And XH-A1 is not even a sound recorder...

Thus "16 is more like 12-14 bits" is a bad genaralisation. Good recorder gives at least 15 bit resolution at 16, 19 bits at 24.
 
Petrus said:
Most (all?) 24 bit recorders use always 24 bit ADC and dither down to 16 bits when recording at 16 bits. That means that it as good 16 bits as possible (if 24 bit ADC is anything better than 16 bits, that is)..
Click to expand...

That's not what I mean. What i mean is that when recording at 24-bit you never really get to use all 24 bits. When you record on a 16-bit system you never really get to use all 16 bits. You wouldn't unless you're recording at FULL LEVEL ALL THE TIME, and that doesn't happen. As a result, most 16-bit recordings are really only using 12-14 bits.

Regards,

Ty Ford
 
First of all, Merry Xmas to all!

Noiz2 said:
As I said, depends on how you define it. The stream IS destructively changed at each level point. There is no effective difference between altering the stream an saving the altered file and then processing it again with the next level change. Also YOU may not think it's a big deal but jump over to some professional audio boards and the issue of mixing and math is HUGE.
SK
Click to expand...

I'm not talking just out of experience with camcorder ADs. I have been writing dsp code, simulation and implementations in plugin format and hardware for quite a few years. I have written mastering floating point and integer software from scratch and I have done research in this field. I have spend thousands of hours discussing dsp with the worlds greatest mastering engineers. I was there when the first mastering solutions appeared and followed the progress and all the errors and bugs of various mixing engines in various platforms, hardware and native. I have studied the implementation of all major audio editors. In simple words, I'm not talking out of my head when it comes to dsp:)

A long series of multiplications (volume changes) in your floating point native mixing app has no audible effect in the sound and is usually not even measurable with analysis. You would be suprised of how many multiplications and adding the signal goes through when you do a simple one band EQ using IIR (a theoreticaly Infinite Impulse Response tail) and that's also when it's not doing any real filtering (flat=0dB). Can you hear a single band of (non oversampled) IIR eq when it is set to flat? Possible on a terrible integer EQ written by an 8yo. Can you hear a single high precision multiplication? Not likely.

Saving a file between processing truncates the signal to 16bit or whatever the editor uses. A real time application of effects does not go through these integer math conversions all the time, it stays in high precision float, so it has a different output. You can test this by two simple volume changes. Take a sine wave signal, do -60dB, save, open, do +60db, check the waveform. Real time output in a single step will have no problems, the output with the save in between could have serious aliasing depending on implementation and format.

24bit to 16bit is usually not massively different in noise, sometimes even alias. On a Sound Devices AD I have here, 16bit difference to 24bit in A weighted dynamic range is only 7dB. A little more than a bit. Not the 8bits one would expect. Add to that the antialias effects of the thernal noise of the mics/preamp and its high level, and the practical difference in a well engineered recording with proper levels could be zero. A 24bit AD rocks in 16bit though, so the 24bit could be massively better than an aging 16bit AD in a typical camcorder (Sony VX series? :) ). In this case we would be comparing different ADs though which is not fair. The 24bit external recorder upgrade is an impressive upgrade but that's only because it is a new design with good engineering and a high price tag, not because of 24bit capability. Its 16bit mode would also be better that the high school project engineering quality circuit on the camcorder.
 
Last edited:
Otis,

I'm not sure anyone questioned your experience and we basically agree. You are making the point I started which disputed a claim by another poster that any change in , what he called "format", created distortion. When I talked about "saving the file" I meant in the processing format not the original format. Yes if you output at every step there would be a huge drop in quality. My point was and is that the stream passing through is changed in "real time" destructively by each step. I didn't mean that the steps caused audible problems but just the opposite, that as you say nobody is going to hear it.

As to the math thing, I'll bow to your knowledge on that. But the math of for instance ProTools mixer section was a long and raging debate. And as I understand it Pro Tools HD uses 48 bit fixed representations during processing and Le uses 32 bit floating. But you said floating point math, and I assume folks haven't been using integer math for a long time, but???
 
On interger you have to watch your levels. Integer math is tricky, especially single precision 24bit. It has to do with how numbers are represented in the format vs floating point. Simply put, floating point has the right distribution on its number axis.

We don't have to worry about 48bit precision integer implementations though in 2008. Not for single processes at least. The implementations are mature and the precision is high. I do understand why users are sceptical. The first years were pretty bad. In greek we have a saying which translates to "if you get burned in hot porridge, you make sure your yogurt is cold enough" :)

Floating point is a natural format for computers and native processing and almost all native editors use that. The computer FPUs have excellent precision and speed in these calculations. Using integer in native processing would be counterproductive and problematic. Which translates to an expensive product with more issues.

Integer is how things were done in hardware for many years and many professional platforms (and some good outboard) still use integer DSPs. The high precision (48bit) combined with the maturity in software engineering solved the quality issues. There is also a huge investment in the engineering behind the algorithms and a wide availability of integer DSPs. Not to mention the intellectual property issues that have to do with the status of the market and the advantages of the hardware based dsp in terms of piracy and profitability.

Generally, we don't have to worry or debate on integer vs float math. All editors will do the job just fine.

There are some terrible dsp implementations on the market (older plugins), usually bundled. Most users have heard about these issues or tested themselves, so they avoid them.

Since LE is using an FPU, it is technically double precision (64bit float) like most native processing apps. Some processing apps can send double precision numbers from one dsp process to another, so they are technically superior to those who use single precision (32bit float) in the links between processes, but don't expect audible differences. The dsp is still done in double precision 64bit float where it counts (inside a dsp process, in IIR tails, signal detection sidechains in dynamics etc).
 
Sound devices on their comparison page use a -40dB signal level AND 128kbit mp3 compression on top in order to show the differences of 24bit and 16bit:

http://www.sounddevices.com/notes/recorders/real-world-24-bits/

Cineform use the same technique here:

http://indiefilmlive.blogspot.com/2005/11/comparing-10-bit-to-8-bit.html

Do I see a pattern here? :) In order to show a difference they have to magnify the difference using a very unrealistic scenario. Extreme signal level deterioration AND low precision storage that is immediately undone (just to truncate) or low level recording (to record low signal quality and boost to show how bad it is supposed to be).
 
Otis Grapsas said:
...
Do I see a pattern here? :) In order to show a difference they have to magnify the difference using a very unrealistic scenario. Extreme signal level deterioration AND low precision storage that is immediately undone (just to truncate) or low level recording (to record low signal quality and boost to show how bad it is supposed to be).
Click to expand...

I won't argue the science/ engineering, but I can talk about the practical side in the field. Our "tests" didn't involve any high tech scopes etc. We simply recorded a variety of sounds and different SR and Bit depths and brought them all into a PT 96/24 session and played them to a bunch of people who didn't know which clip was which. Everybody liked the 24 bit recordings better. 96 VS 48 a lot of people couldn't tell the difference. Admittedly these were sound people and the general public might not have picked up on the 24 bit so much? So what ever the reason is, in practical terms I will always record 24 if possible because I can hear the difference. Now maybe I'm technically just getting a really good 16 bits? But when recording at 16 bits I'm not getting a really good 16??? Either way the end result is that recording at 24 gives me a better sounding recording. And I have found this to be true on a fairly wide range of recorders ~ $800 - $7,500.

cheers
now go have some eggnog!
sk
 
The test was using two copies of the same converter/recorder operating in 16bit and 24bit recording the same sound? How did you handle microphones and signal? Comparing using a good methodology should preserve the recording character and only show a difference in alias/noise. On a typical theater level, these differences should be well masked by the environment. Processing can bring them up to a degree though.

From tests on the subject, even a significant difference in alias/noise will not be perceived directly on a single subject, but on multiple trials and multple subjects, in other words, statistically. The fact that everybody liked it is very strange by itself! An ABX test (the correct way to free tests from errors) should be quite accurate if you get the chance.

I'm not very young at 33. My ears are probably average although I try to protect them (never worked profesionally in loud environments, I avoid clubs, monitor at medium levels, etc). I think I can hear a small difference between 16bit and 24bit recordings of analog material, but I don't think it's something I would be able to notice with the presence of a moving image. I might even fail an ABX comparison:)

Sampling rate differences are very hard to perceive. Many tests reached that conclusion. Actually, most tests that actually showed a difference was very unscientific and sponsored:)

Any decent 24bit should be quite good at 16bit and 1-5 bits better than 16bit in 24bit. There is no reason not to use 24bit if it is available. More is better in this case. Even an extra 6 to 12dB is helpful for preserving quality while having additional headroom or avoiding a limiter.
 
tyreeford said:
That's not what I mean. What i mean is that when recording at 24-bit you never really get to use all 24 bits. When you record on a 16-bit system you never really get to use all 16 bits. You wouldn't unless you're recording at FULL LEVEL ALL THE TIME, and that doesn't happen. As a result, most 16-bit recordings are really only using 12-14 bits.
Click to expand...

If someone is recording with -24 dBFS peaks it is user mistake, not system failure.

Just the other day I recorded an organ concert where the first 4 pieces were really quiet, around -36 dBFS. Now, I should have had a 32bit recorder for that, are my takes ruined for lack of bits? Blower noise is around -62 dBFS, by the way.

I do not see any point in complaining that on normal 16 bit recordings the average level hovers around 12-14 bits used. With normal listening levels you have to adjust the listening volume for 15-16 bit peaks, 12-14 bits level sound just nice and still the noise levels are somewhere far below ambinet noise and often even below hearing limits for most reproduction systems and rooms.
 
Petrus said:
If someone is recording with -24 dBFS peaks it is user mistake, not system failure.

Just the other day I recorded an organ concert where the first 4 pieces were really quiet, around -36 dBFS. Now, I should have had a 32bit recorder for that, are my takes ruined for lack of bits? Blower noise is around -62 dBFS, by the way.

I do not see any point in complaining that on normal 16 bit recordings the average level hovers around 12-14 bits used. With normal listening levels you have to adjust the listening volume for 15-16 bit peaks, 12-14 bits level sound just nice and still the noise levels are somewhere far below ambinet noise and often even below hearing limits for most reproduction systems and rooms.
Click to expand...

Jesus, Petrus, you're one contentious guy. I'm not complaining and i'm not talking about 32-bit. Stay on the subject will you?

I'm stating fact. 20-bit recordings from 24-bit machines are better than 14- bit recordings from 16-bit machines, all things being equal. I have little problem with a 16-bit master, but having 24-bit in production makes a difference.

Ty Ford
 
tyreeford said:
I'm stating fact. 20-bit recordings from 24-bit machines are better than 14- bit recordings from 16-bit machines, all things being equal. I have little problem with a 16-bit master, but having 24-bit in production makes a difference. Ty Ford
Click to expand...

You missed my point. Ferrari with a bad driver is slower than KIA with a good one . When used properly 16 bit is plenty good enough for almost everything, and always for final delivery (practically nobody has system and rooms where difference could be heard). For processing and recording the original 24 bit has advantages.

Peace on earth...
 
No, I'm decidedly not missing your point, I'm disagreeing with your ability to be on both side of the issue within the same post.

Ty Ford
 
We know 24bit is better but we also know 16bit is good enough if used with care because there is overwhelming evidence. There is no contradiction there.

Someone interested in a movie or a documentary might stop and say: It's DV isn't it? I do need HD for distribution because my channel has this requirement. But who will notice audio that started with 16bit recordings or give a damn?
 
You must log in or register to reply here.
Back
Top