Truncation: Inaudible or Significant ???

  • Thread starter Thread starter mark4man
  • Start date Start date
mark4man

mark4man

MoonMix Studios
Here's a tech question for all you mastering types out there:

I am right smack in the middle of a process c/o bouncing track groups & comp sets from my 16-Bit projects (which have more than a fair amount of DSP goin' on...plug-ins, automation envelopes, level/gain changes, etc.) to 24-Bit stems...in preparation of then taking those stems to the mastering house.

I always assumed that 24-Bit BTD's would grab the extra word length associated w/ the DSP commands & plug-in ambience.

Then I opened up a few of those stems in WaveLab; & found that some were maxed out at a full 24-Bits of sonic activity.

So now...I'm worried about truncation; & the effects of it.

There seems to be two schools of thought on this...the first from Bob Katz:

Inside a digital mixing console (or workstation), the mix buss must be much longer than 16 bits, because adding two (or more) 16-bit samples together and multiplying by a coefficient (the level of the master fader is one such coefficient) can result in a 32-bit (or larger) sample, with every little bit significant. Since the AES/EBU standard can carry up to 24-bits, it is practical to take the 32-bit word, round it down to 24 bits, then send the result to the outside world, which could be a 24-bit storage device (or another processor). The next processor in line may have an internal wordlength of 32 or more bits, but before output it must round the precision back to 24 bits. The result is a slowly cumulating error in the least significant bit(s) from process to process. Fortunately, the least significant bit of a 24-bit word is 144 dB down, and most sane people recognize that degree of error to be inaudible.
Click to expand...
The 2nd is from Ken Pohlmann:

Discarded data results in an error analogous to that of A/D quantization.
Click to expand...
The implications of this is that the distortion would be spread across the entire bit scale, correct?

So...who's right...if we whack off a 24-Bit file at 16 bits...is the damage only to the sonic info at the base of the 16-Bit floor? Or is it spread throughout the entire audible range? [Or...is Katz meaning rounding as "rounding"...& if so...what's the difference (rounding vs. truncation)]?

Also: In my above scenario...If the internal precision of the workstation is 32 Bits or greater (which all are)...isn't that also truncation, to pipeline the signal to 24-Bits (to both the interface for playback & also BTD)...& as such...requires dithering at that point?

Thanks,

mark4man
 
They're both right, just saying different things. Bk's quote has to do with truncation at 24/32 bits not 24 to 16 bits.

The effect is more noticable at lower levels. You can consider it as a ratio of the of the level of distorted audio to undistorted. However I feel the affect is still very noticeable at higher levels. Truncation gives audio a grainier colder sound than audio that has been dithered properly.

To answer your question, If I were your ME I would suggest that you send the original 24 bit or 32 bit float files rather than a truncated or dithered version and process the files at 24 or 32. Part of the reason is given in by your BK quote above. Less distortion if processing 24 bit files "down the line" than changing wordlength through the chain.
 
I always dither my tunes, but in all honesty, I prefer the truncated sound. The dithered version sounds fuzzy to me. I must be a freak of nature or something :confused:
 
mshilarious said:
I always dither my tunes, but in all honesty, I prefer the truncated sound. The dithered version sounds fuzzy to me. I must be a freak of nature or something :confused:
Click to expand...

What types of dither have you been using? Some are more transparent than others, and noise shaping is also a variable.

Dither in general has a slightly veiled sound, but it should be warmer than truncation.
 
mshilarious said:
I always dither my tunes, but in all honesty, I prefer the truncated sound. The dithered version sounds fuzzy to me. I must be a freak of nature or something :confused:
Click to expand...
I also think this another one of those genre-specific things.

G.
 
SouthSIDE Glen said:
I also think this another one of those genre-specific things.

G.
Click to expand...

I respectfully have to disagree there G. I think what you guys are hearing is an artificial brightness to the material. The sound is similar to when you are clipping the converters a bit, probably due to aliasing.

It's better to EQ this in and have control rather than rely on truncation IMHO. Of course some people clip purposely and otherwise distort, but again I think there are better ways of acheiving the same effect.
 
masteringhouse said:
I respectfully have to disagree there G. I think what you guys are hearing is an artificial brightness to the material. The sound is similar to when you are clipping the converters a bit, probably due to aliasing.
Click to expand...
I'm with you completly, Tom. I probably should have explained what I meant better.

There's no question that great dithering does a great job of keeping the apparent sound "equalized" from pre-dither to post-dither, keeping a seemingly accurate and natural character to the sound. Truncation, OTOH, artifacts the sound in a very unnatural and seemingly less accurate way. I'll take dithering over truncation any day, regardless of genre.

But the longer I frequent this forum, the more I'm understanding where a large percentage of other visitors to this forum are coming from. I'm not saying I agree with them, I'm just saying I understand them.

There is a largely disproportionate number of folks on this board who are almost exclusively into the heaver isotopes of rock and/or an engineering style where sensationalist engineering tricks are just as - if not sometimes more - important to their recording as the actual musical performance is. Any and every chance they can get to hard pan, distort, turn to eleven, and any other way to get a "k3wl" sound to augment their recording, they'll take. The further from "normal", the better (it's all part of the angst/rebellion thing.) In such a realm, I can easily understand someone thinking the harsh realities of truncation or junk food dithering might actually be a positive thing.

Do I agree? No, on a couple of different levels. But can I understand why others might like the sound of truncation? Well, maybe I can't understand why they think it sounds good, but I can understand that they could think it sounds good anyway.

G.
 
SouthSIDE Glen said:
Any and every chance they can get to hard pan, distort, turn to eleven, and any other way to get a "k3wl" sound to augment their recording, they'll take. The further from "normal", the better (it's all part of the angst/rebellion thing.) In such a realm, I can easily understand someone thinking the harsh realities of truncation or junk food dithering might actually be a positive thing.
G.
Click to expand...
Isn't this harsh environment also a reason that certain music genres just turn out to be fads? Eventualy, don't peeps grow out of these styles because their bodies(ears) begin to reject the loud and harsh sounds just like the junk food?
 
NYMorningstar said:
Isn't this harsh environment also a reason that certain music genres just turn out to be fads? Eventualy, don't peeps grow out of these styles because their bodies(ears) begin to reject the loud and harsh sounds just like the junk food?
Click to expand...
One would hope so :). And I think for the most part that happens. But there are a couple of things that mitigate that.

First there is a subset of folks who simply find a genre they like early on and stick with it. This Friday I'm working a biker bar where the average age of customer is probably around 35 yrs old. Yet before and after the band comes on (we've played there many times before) they are pumping quarters into a juke box with the very lightest music to come out of it being somewhere in between Nickelback and The Scorpions.

Second is that for every person who does grow older and whose tastes evolve, a new young whippersnapper comes along all filled with the same angst and rebellion that the old man used to have and re-fuels the genre. You could probably say that headbangers are a renewable resource :).

Hmmm...if we could turn long hair, wallet chains and heavy distortion into fuel... :D

G.
 
Thanks masteringhouse (& everyone)...

Great responses...all.
(sorta gettin' a bit sidetracked...but...HEY...I don't recall my placing any restrictions on thread content; & most of it's really very interesting anyway.)

But...

masteringhouse said:
To answer your question, If I were your ME I would suggest that you send the original 24 bit or 32 bit float files rather than a truncated or dithered version and process the files at 24 or 32. Part of the reason is given in by your BK quote above. Less distortion if processing 24 bit files "down the line" than changing wordlength through the chain.
Click to expand...
That reminds me...I think you may be right (&...you may have kept me from f_ck_n' up big time.)

As I said earlier...my tracks began their lives at 16-Bits, underwent various DSP; &...in bouncing sets & comps to 24-Bit stems...presumably everything was captured (& I checked a few of them on the Bit Meter in WaveLab...& none of them exceeded 24-Bits of sonic activity.) The idea was for the ME to open those in his (floating point) mix environment & output from there...(he's going the "push the output converters...clip the input converters" route, since any type of compression/limiting really screws up my sound.) For these stems...they're OK as is...each one is 24-Bit; & as a composition group, they'll be output from a floating point environment & recaptured at 24/44.1...returned to me at 24/44.1, where they'll undergo TPDF back to 16 & I'm ready.

But (& this is my main concern)...all of the 24-Bit stems were opened in 24-Bit master projects just to hear what they sounded like. In some cases, I had to perform some secondary tweaks & even some "touch-up" DSP (to those master projects.) Now, I can't bring those to the mastering house & sit there & have the ME apply automation envelopes or touch-up processing, so for those...he's having me bounce them to stereo files.

Since those master projects are made up full 24-Bit stems; & since there's undoubtedly more going on than 24-Bits at the outputs of those projects...even if only from cumulative calcs, let alone the extra DSP...

...instead of dithering those outputs to capture everything in the stereo bounce (which I almost did; & the session is this coming Wednesday)...whereby my finished masters will wind up with two instances of dither (not good)...is this why it's better to bring 32-Bit float stereo .wavs?

[Also...If I bring 32-Bit Floaters...I don't have to turn down the master faders to stay under 0dB, correct? So...he won't have to implement make-up gain at his output end, right?

(Sorry to take so long to get to the point...but I wanted to get the fact correct (since I'm a digital musician & not an ME or audiophile.)

What thinks y' peeples 'bout this here?

mark4man


O, BTW - On the truncation thing...although, as said, grainyness & artificial brightness do result...sometimes it adds just the right amount of "sizzle".
 
Just stop.

4man, I can't believe you are still chasing this dam thing. You done pissed off a bunch of people at PSW with your craziness. Just take your OCD pills and get this thing done. You are being far too protective of your little baby. 32-bit float is a crutch for those who don't have sense enough to keep things at a reasonable level in their little software mixer. It is not giving you magic sound. It is not saving you the "detrimental effects" of makeup gain. Until you find some 32-bit convertors, don't worry about it. If you are so worried about dither and such, just play your project out through the mastering chain (with a hardware digital limiter on the end) and recapture at 44.1K 16-bit to be burnt to CD. Once your audio makes the conversion from D to A (assuming at least a partly analog mastering chain), it no longer has a bitrate and dither is no longer needed. So get on with it already.

There, someone had to say it.
 
OK, I'm confused... when do I dither? Should I dither mixing down to 16-bit? 24-bit? should I try both and listen to the mixes back to back to and see if I hear a difference? (yeah I probably should...)

I'm assuming that all host applications' dithering is not created equal. or is it?

(sorry, I don't mean to hi-jack but this seemed like a good place to ask about this!)
 
masteringhouse said:
What types of dither have you been using? Some are more transparent than others, and noise shaping is also a variable.
Click to expand...

Apogee UV22HR, usually set to Low and Autoblack. Genre is often acoustic stuff or Pop/Gospel chorale, final RMS maybe -18 to -20 which means there is a fair amount of emptyish space.

Don't get me wrong, I ain't gonna stop dithering, I'll just trust the math rather than my ears :confused: :o
 
mshilarious said:
I always dither my tunes, but in all honesty, I prefer the truncated sound. The dithered version sounds fuzzy to me. I must be a freak of nature or something :confused:
Click to expand...

For a very long time I preferred to not dither at all. I just felt like dithering was really just costing me an extra bit, and simply caused your 16 bit file to actually contain only 15 bits of relavent data.

That was until I got a converter that does outboard dithering...it made all of the difference in the world.

Anywayz, back to the whole 'the noise is at -144 db so its inaudible'....it is my own line of thinking that the last bit will be just as present in the louder material as in the quiet material. why?? the last bit does not just define the audio at -144 db. it defines the audio at ALL db..it definies the most minute change in decibel level. it is the difference between 0 db and .99999 db. it is always there, at all volume levels. it is not simply

00000000 00000000 00000001

it is also

11111111 11111111 11111111

get what I mean?
 
Reggie...

Who the hell are you & what difference does it make to you, anyway?

I don't see anyone around here getting "pissed off" save for you; & PSW was a different issue (which has been amicably resolved.)

Look...In my inexperience, I started off a few years back w/ shitty 16-Bit recordings & I've been trying to make them sound as good as possible ever since. It happens to be important to me that I release the best sounding (first) project I can, so I can then go on to the next project knowing I did everything I could to achieve that goal.

If that doesn't happen to sit well with you...go jump in a lake (& take your insults with you.)

mark4man


BTW-& 32-Bit float is not a crutch...most of the major DAW's offer that export option now; & more than a few ME's regard it as a viable option.
 
Reggie said:
32-bit float is a crutch for those who don't have sense enough to keep things at a reasonable level in their little software mixer. It is not giving you magic sound. It is not saving you the "detrimental effects" of makeup gain. Until you find some 32-bit convertors, don't worry about it.
Click to expand...

32-bit float is far from a crutch....

All audio processing in nearly any DAW (and in many operating systems themselves) is done in 32-bit float. This has little to do with not being able to keep things at a reasonable level. Indeed, the maximum dynamic range of 32-bit float is pretty similar to 24-bit signed integer.

The reason that the 32-bit float format is used is that your computer naturally moves data around in 32-bit quantities. As such, it is an easy quantity to fetch from and send to memory. Floating point values are also easier to work with when doing various types of audio math in plug-ins.

Floating point numbers are also a convenient size to work with in vector processing, which can be very useful for accelerating audio transforms, etc.

Finally, floating point is more suitable for mixing, as it allows intermediate results to exceed 0 dB with impunity even if the final result will be within bounds. In extreme cases, this can make a significant difference in sound quality, since otherwise you would have to mix at a significantly lower level to avoid clipping.
 
masteringhouse said:
A good read:

http://www.rane.com/note153.html
Click to expand...

A few points after reading that:

Right off he states that there is probably little practical difference in properly implemented fixed or floating point. Then he goes on to trash floating for the rest of the article :rolleyes: But never really explains why floating really probably is OK :confused:

When he discusses the pros and cons of floating, essentially he states that it's easier to program, whereas there seems to be some risk, if not quantified, that fixed could be screwed up by the programmer. Certainly that would have to be weighed by a developer, along with issues such as heat dissipation. However, for a DAW software company, the physical problems of floating-point design have already been solved for them by Intel, AMD, et al. Therefore they simply have a set of specifications for CPU power they must consider.

At the same time, the risk of improper fixed-point design seems to fall entirely on them. Faced with a larger number of hours required to design a proper fixed point DAW, and the resulting implications on product pricing, it's quite clear that such considerations are not trivial, at least compared with power consumption.

From a practical point of view, he makes certain claims about the performance of floating point vs. fixed when mixing signals, rather than just performing gain change operations that cause transient overs. I've tested the floating-point gain change performance before, but not the mixing bit, so here goes:

I took a single track that peaked around -9dBFS, and mixed it with itself. The resulting file was around -3dBFS. Next, I normalized that file, and compared it with the original, also normalized, and phase inverted.

Now, here is the trouble. I believe his conclusion is that floating point doesn't offer the same precision as fixed when adding integers. But on a practical level, either is better than the level of precision offered by a DAW interface. I can only set gain changes to three significant digits, that is .001dB increments. However that corresponds to something like a -80dBFS signal.

And indeed, that is what the null file contains--a random signal with data in the 15th and 16th bits. Is that because floating point is broken, or because Wavelab won't let me set a gain change of -6.020599913dB :confused:

It reminded me of the sound of dither--except Apogee's UV22HR, set on Low and applied to a blank file, has data in the 14th bit.

Is floating-point therefore self-dithering? :D Discuss ;)

Or maybe it provides that "analog" warmth :p
 
MS -

I don't understand the relevance of power consumption or heat dissipation. The point made by the article is in the way floating point and fixed perform calculations.

In summary the point that I believe he is making is that floating point inherently has 24 bits worth of precision no matter what range of audio you are trying to represent, while fixed point calculations are usually performed at a higher 48 bit precision and then brought back to 24 bit.

The design of the application is key. Some fixed plugs will dither back to 24 bits after calculations in 48 bit, others don't. It's my opinion that those that truncate sound colder when doing this. IMHO all fixed plugs that work this way should at least have an option to turn on dithering, Waves plugs do this. It's also the reason that I use the dithered mixer in a Pro Tools TDM rig.

In the case of floating point calcs, the article sugggests that there are innacuracies once the precision does beyond the 24th bit and can result in a pumping noise floor. Check out the Moorer article for more info on this.

Indeed, when we compare 32-bit floating point with 24-bit integer, there is some advantage to the floating-point. The resulting noise floor for 24-bit audio processing is of a constant level. It does generally not depend on the level of the signal. At low levels of audio, the noise floor may not be far below the level of the audio itself. In floating-point, the noise floor is directly related to the amplitude of the audio itself. If the audio is at a low level, the noise floor will be at a correspondingly lower level. This is generally a good thing, but there is some disadvantage since it is necessarily a changing, or ?pumping? noise floor. The hope is that said noise floor is sufficiently low to be inaudible under all circumstances.
Click to expand...

Is it audible? I'll let others decide that.

In regards to using float as a crutch, I believe the point there was that some budding engineers do not pay proper attention to gain staging and rely on 32 float processing to cover overs rather than making the proper adjustments to fit within the contraints of 24 bit. If you consider that there may be some additional inaccuracy when processing beyond 24 bits in a 32 bit floating point register, it may have some theoretical merit.
 
You must log in or register to reply here.
Back
Top