| Author
|
Is there any demand for 24-bit WAV files?
|
pantelhs
Started Topics :
4
Posts :
11
Posted : Jan 24, 2008 14:58
|
So, from what I have understand it's better to process the sound at 32 bit and dither to 24 or 16 for playback. But, is this also applying to sample rate? I mean processing the sound at higher sample rates and dithering to 44.1 for playback or the loss from dithering is bigger than writing, from the start at 44.1. Or it doesn't matter anyway?
Great thread indeed. |
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Jan 24, 2008 15:29
|
You don't need to dither when going from 32 to 24 bit, only when going from 32 or 24 to 16.
Dithering has noting to do with samplerate conversion.
Working in higher samplerates and then convert it down at the mastering stage can be beneficial if you use plugins without internal oversampling.
But unless you have a lot more power in your DAW than you will need the cost by far outweighs any benefits.
 (``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
Upavas
Upavas
Started Topics :
150
Posts :
3315
Posted : Jan 24, 2008 21:12
|
|
phobium
Phobium
Started Topics :
14
Posts :
718
Posted : Jan 24, 2008 22:52
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Jan 24, 2008 23:44
|
Quote:
|
On 2008-01-24 15:29, Spindrift wrote:
You don't need to dither when going from 32 to 24 bit, only when going from 32 or 24 to 16.
|
|
Opinions differ on that. The main technical argument that I've read against dithering to 24 bits is that you need dither that accounts for all the possible equivalent bits of a 32 bit system to do it properly. (I think that is about 256 bits worth of dither which needs quite a bit more processing power than your regular 8 bits dither from 24 to 16).
Others argue that partial dithering is better than no dithering at all.
Another argument against dithering from 32 to 24 bits is that the quantization distortion without dithering is so low that it would not be audible. Mostly a signal shouldn't go from 32 to 24 bits many times. (Maybe once for the 24 bit mix export and then straight down to 16 bits during mastering). So there is not much cumulative distortion. That is unless you use a ProTools TDM system with many RTAS plugins.
Another argument against is that most typical signals (analogue sources) have such high noise floors compared to 24 bits that the signal is already partially dithered. (But not completely).
Some DAWs don't dither to 24 bits. (Cubase and Nuendo don't unless something has changed since version 2 that I am not aware of). Some DAWs like Sonar do. ProTools leaves that up to the plugin developer for inserts (All TDM plugins receive and deliver 24 bit data. All RTAS plugins receive and deliver 24 bits and use 32 or 64 bits internally unless two RTAS plugins are chained in which case they pass 32 bit data between each other) and lets the user decide for the mix buses (from 48 to 24 bits) by enabling or disabling the dithered mixer plugin.
As for the Boogeyman, he lives in fear. 
UnderTow |
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Jan 25, 2008 01:16
|
I suspected that there might be different opinions on that since I have seen the option in some software to dither to 24 bits.
I just recall from my sound engineering studies that it back then was said that it is not needed and that does seem logical considering that the bit depth of the actual audio signal is the same.
Also isn't it still a no no to apply dithering several times?
And usually you would apply dithering again when going to 16 bit if the audio is destined to be printed onto a CD.
Anyway, thanks for that clarification. I guess that if there is discussions about that it might be beneficial to dither when going from 32 to 24 ideally one would simply stick to 32 bit for recording.
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Jan 25, 2008 11:23
|
Quote:
|
On 2008-01-25 01:16, Spindrift wrote:
I just recall from my sound engineering studies that it back then was said that it is not needed and that does seem logical considering that the bit depth of the actual audio signal is the same.
|
|
Only if the exponent is at unity. Otherwise the 24 bit mantissa doesn't map directly to the bits of a 24 bit signal. And that is the problem: Because of the exponent, the possible range of a floating point format is allot of bits in a fixed point format.
Quote:
|
Also isn't it still a no no to apply dithering several times?
|
|
No. The rule is pretty simple: Dither any time that a bit reduction occurs. That means after any operation that increases the word length. Even something as simple as a gain change should be dithered when the results of the multiplication gets truncated back to the original word length.
A big no no is to use shaped dither several times. That should be left for the very last bit reduction to 16 bits.
Quote:
|
Anyway, thanks for that clarification. I guess that if there is discussions about that it might be beneficial to dither when going from 32 to 24 ideally one would simply stick to 32 bit for recording.
|
|
Well AD converters only give about 21 bits of real signal. Below that is just noise from the electronics, mics, preamps etc. And most converters give less than this. 24 bits is fine for recording. 32 bits and above is good for processing to keep the resultant numbers as precise as long as possible.
UnderTow |
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Jan 25, 2008 13:19
|
Quote:
|
On 2008-01-25 11:23, UnderTow wrote:
Well AD converters only give about 21 bits of real signal. Below that is just noise from the electronics, mics, preamps etc. And most converters give less than this. 24 bits is fine for recording. 32 bits and above is good for processing to keep the resultant numbers as precise as long as possible.
UnderTow
|
|
Ok, but you are saying that conversion from 32 to 24 bit is not transparent.
So if you record all your tracks as 24 bit you would either have distortion buildup if you don't dither or noise buildup if you do...no?
If the noise or distortion is at a lower level than any DAC can recreate, even when adding up from several tracks, I can see that there is no advantage of using 32 bit for recording (apart from being able to be a bit more lax with your levels), but then I guess it doesn't matter if you dither or not either?
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
Trip-
IsraTrance Team
Started Topics :
101
Posts :
3239
Posted : Jan 26, 2008 01:16
|
Gentlemen,
With all due respect - my daw doesn't "record" 32bit floating point. Recording is done at the usual 24bit, as the adc implies, the daw then adds those 8 bits for internal processing. With a 16bit file it'll add 8 more bits (zeroed) - 24bit conversion, and more 8 for the 32 bit.
What I'm saying is that, I don't get to choose working IN the daw wether recording 24 or 32. The bare phrase 'recording file at floating point bitrate' doesn't sound "Logical" to me (  ) - everything is always 32bit floating point while in the daw.
When I bounce a finished project to 24bit, or yet pre-mastering, I don't apply diether (POW-r) - though I can. But it never occured to me to apply dithering before further processing. I suppose one-time truncation from 32 to 24 without dithering is inaudible, but maybe repeated truncating when reimporting into the daw might become audible.
Quote:
| On 2008-01-25 11:23, UnderTow wrote:
No. The rule is pretty simple: Dither any time that a bit reduction occurs. That means after any operation that increases the word length. Even something as simple as a gain change should be dithered when the results of the multiplication gets truncated back to the original word length.
A big no no is to use shaped dither several times. That should be left for the very last bit reduction to 16 bits. |
|
and here's a reading marathon:
http://www.soundonsound.com/forum/showflat.php?Cat=&Number=552666&Main=537295
Crackling universes dive into their own neverending crackle...
AgalactiA |
|
|
|
Basilisk
IsraTrance Full Member
Started Topics :
168
Posts :
2984
Posted : Jan 28, 2008 03:35
|
It seems as if there aren't any clear answers in this discussion. I haven't personally heard any convincing tests one way or another and don't plan on paying much attention to 24-bit quality (or higher sampling rates) until I make the switch to laptop DJing.
As far as public interest goes, I was able to test the waters with the recent release of 55 Full Moons, the most recent compilation from Cosmogenesis Recordings. By request of the label, this release is available in 320k MP3, 16-bit WAV, and 24-bit WAV. Here are the current download counts for each package...
MP3: 836
16-Bit WAV: 324
24-Bit WAV: 54
Total: 1,215
Few people seem interested in the 24-bit WAV package but there is some demand at least... and this is only after 5 days of tracking. Well, it's food for thought... but I doubt I will be encouraging any content providers to make 24-bit versions available anytime soon.
|
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Jan 28, 2008 13:58
|
Quote:
|
On 2008-01-26 01:16, Trip- wrote:
Gentlemen,
With all due respect - my daw doesn't "record" 32bit floating point.
|
|
No but some DAWs do allow it. (Pointless IMO but still, it is possible). So the question is worth asking.
Quote:
|
When I bounce a finished project to 24bit, or yet pre-mastering, I don't apply diether (POW-r) - though I can. But it never occured to me to apply dithering before further processing.
|
|
POW-R wouldn't be a good choice for this as it is a noise shaping dither. If there is any subsequent processing, TPDF (Triangular) dither should be used.
Quote:
|
I suppose one-time truncation from 32 to 24 without dithering is inaudible, but maybe repeated truncating when reimporting into the daw might become audible.
|
|
The general idea is that truncation distortion is more objectionable to the ear than noise. Also, truncation distortion peaks higher than dither noise.
I don't know if it is audible (I havn't noticed a difference in my setup) but it is included as an option in my DAW and doesn't cost anything to turn on.
That might not be the best source of information (I haven't read it). I distinctly remember explaining to Hugh Robjohns why the Roger Nichols article on 16 vs 24 bit audio was seriously flawed. It took some time for him to understand the technical inaccuracies in the article.
This article by Nika Aldrich is a very good start: http://www.cadenzarecording.com/dither.html
UnderTow |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Jan 28, 2008 14:03
|
[quote]
On 2008-01-25 13:19, Spindrift wrote:
Quote:
|
Ok, but you are saying that conversion from 32 to 24 bit is not transparent.
|
|
I'm not saying it isn't transparent. I'm saying that the bits of the 32 bit mantissa are only directly equivalent in a 24 bit word when the exponent is 0. (Or was it 1?). If the level goes up or down and the exponent changes, there will be truncation (or clipping) when the 32 bit word is converted to 24 bits.
Quote:
|
I can see that there is no advantage of using 32 bit for recording
|
|
As Trip- pointed out, there is a difference between recording 32 bits to disk (not useful) and processing at 32 bits (useful).
UnderTow |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Jan 28, 2008 14:05
|
Quote:
|
On 2008-01-28 03:35, Basilisk wrote:
MP3: 836
16-Bit WAV: 324
24-Bit WAV: 54
Total: 1,215
Few people seem interested in the 24-bit WAV package but there is some demand at least...
|
|
Ah real numbers, great! Still, I suspect some people will just select the 24 bits files simply because they are available (without really knowing the implications). I doubt there would be any complaints if there where only 16 bit files available which is the real test IMO.
UnderTow |
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Jan 28, 2008 15:20
|
Quote:
|
On 2008-01-28 14:03, UnderTow wrote:
I'm not saying it isn't transparent. I'm saying that the bits of the 32 bit mantissa are only directly equivalent in a 24 bit word when the exponent is 0. (Or was it 1?). If the level goes up or down and the exponent changes, there will be truncation (or clipping) when the 32 bit word is converted to 24 bits.
|
|
I'm not sure I understand what you are saying here.
Obviously one has to assume that usually the exponent is not 0, and you are saying that that means that some truncation would occur.
And the reason we like to apply dither is because the truncation is not transparent...isn't that so?
So if the signal does suffer from the 32 to 24 conversion, why not stay in 32 bit when recording?
And if the signal does not suffer from the conversion, then why are we discussing the truncation and dither process at all?
I guess the differences is marginal at best and this can be more of an academic question that one with obvious practical consequences.
But it's nice to avoid signal degradation in every way possible.
And I do think it's nice not to always have to ensure that you get optimum recording level when recording (obviously excluding recording analogue sources when you want to get max resolution from the converters), so recording as 32 bit does have some of the same advantages as when mixing...at least for lazy people like me.
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
UnderTow
Started Topics :
9
Posts :
1448
Posted : Jan 28, 2008 17:18
|
Ah! Now things are making sense: You mean bouncing to disk, rendering or recording to a new track from within the DAW? Then yes, keep the signal at 32 bits. (The term recording is a bit confusing here as it can mean various things).
For recording analogue sources or digital sources from outside the DAW (AES, ADAT, SPDIF etc) then 24 bits is enough.
And yes the difference is marginal at best. But as you say, why degrade the signal when you don't have to?
UnderTow
|
|
|
|
