R/E/P Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: [1]   Go Down

Author Topic: 24 or 32fp in mastering?  (Read 2790 times)

OatBran

  • Newbie
  • *
  • Offline Offline
  • Posts: 21
24 or 32fp in mastering?
« on: May 20, 2004, 04:14:07 pm »

  While I understand the differences between the two, I am curious what the mastering society tends to use for both processing and for capturing.

 There are people who swear by straight 24bit while others say 32bit floating point is much better. I do understand that 32 floating point is less CPU intensive for most processing applications as well and that might be why some gear/apps choose this instead.

 Most current apps allow you to choose which you prefer.  Some only give you 1 option.  I'd like to hear what some of you think on this subject.

Brian
Logged

TotalSonic

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 3728
Re: 24 or 32fp in mastering?
« Reply #1 on: May 20, 2004, 08:30:19 pm »

Main thing to realize is that a 32bit IEEE file is really just a container for only 24bits of audio information - so it does not store any more information than a 24 bit wav or aiff file.  

What the 32bit float file is useful for is extending internal processing math calculations (which happens when you do something like summing multiple tracks, increasing gain, or doing processing such as eq's or reverbs) beyond 24bits.  However most mastering & DAW apps and hardware boxes calculate well beyond 32bits. For example, both Sonic HD & Pro Tools HD use 48bit fixed point math for their internal processing.

The other controversy is the advantage of fixed point "integer" math over floating point math for audio dsp.  A lot of developers believe better results can be achieved using fixed point instead of 32bit float.  An interesting article on this can be read at http://www.rane.com/note153.html

The native PC DAW I use (and resell), SAWStudio, does not save files as 32bit IEEE - but it sounds to my ear superior to many apps that use 32 bit float.  It processes 24bit wav files internally using 32 bit "double precision"  integer math (where, from my limited understanding, each number is represented twice when doing calculations as a method of achieving more accurate results  with many internal calcualations done using 64bit double precision integer math.  The 3rd party eq's and some of the other native processing plugins I use with it calculate using 64bit floating point math within it.  

Best regards,
Steve Berson

OatBran

  • Newbie
  • *
  • Offline Offline
  • Posts: 21
Re: 24 or 32fp in mastering?
« Reply #2 on: May 20, 2004, 10:16:05 pm »

It is the use of the fixed or floating in processing that I am particularly refering to.

I am not sure if PT is fixed or floating.  Might need to check on that.  Though I am sure Sonic is fixed.  I know Steinberg products are all floating and tracks that way as well (unless told to do otherwise).  I know Sonic Foundry/Sony products are fixed.  I am pretty sure Logic is floating though I am not positive on that one either.

I have been hearing debate as to which 'sounds better' and that is the question that I am actually looking to hear opinions on.  I have worked extensively with both and was curious if anybody else had any opinions on the subject.  I am seeing both used in hardware as well (effects, verbs, ect).  Most if not all work at double bit rates (48 or 64bits) internally as you said.

While the sonic differences are very minor, I do believe they are there.
Logged

bblackwood

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 7036
Re: 24 or 32fp in mastering?
« Reply #3 on: May 20, 2004, 10:19:59 pm »

I do no internal processing (save a safety limiter for cross-fades and such), so everything is recorded 24 here with all fades and such processed at 32 float...
Logged
Brad Blackwood
euphonic masters

TotalSonic

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 3728
Re: 24 or 32fp in mastering?
« Reply #4 on: May 20, 2004, 11:01:49 pm »

bblackwood wrote on Fri, 21 May 2004 03:19

I do no internal processing (save a safety limiter for cross-fades and such), so everything is recorded 24 here with all fades and such processed at 32 float...


Makes sense - if you've got great analog hardware like you do there's really no reason to futz with the plugz - and your end results have certainly spoken for the effectiveness of this.
However for things that I like to think of "micro-mastering" where an eq or comp is used only on a single note or phrase that is sticking out in a weird way - or to have the eq change very subtly to reflect the different vibes of the passages of a song (i.e. slightly brightening up a really muted verse but warming up a chorus where the drummer is bashing the crap out of a piercingly recorded crash)-  then the precision and automation (and yeah, the convenience) possibilities found in a DAW can be useful.   i.e. do the major global stuff analog or with the big digital guns and the tweaky detail stuff with the plugz.   Then again I have a feeling that I might deal with stuff that has been poorly recorded/mixed and has no chance in hell of being remixed better where these kinds of things can be more of a necessity than others do - so others mileage will vary.  

Obviously if someone uses a hardware L2 or a Weiss DS-1 or similar digital processor then they are indeed doing internal processing (in the case of the L2 at 48bits) - it's just chips inside a rack box instead of in a computer - but it's math none the less. To date most hardware was able to not chintz on processing power or algorithms like code written for native apps had to (as people coding for native DAW had to make sure that they wouldn't choke the audio stream) but with 3GHz cpu's being a norm now the potential is there for a well written DAW processor to be pretty darn good and perhaps even superior to older digital hardware.  i.e. to my ear my SAWStudio workstation sounds a heckuva lot better than the Neve DTC it sits next to.

As far as floating vs. fixed - this is probably just as  contentious as any argument gets - and I definitely don't know enough about it to really weigh in with anything but conjecture based on limited experience.  Still, I think the differences between the 2 are much more obvious when summing tracks (where in my own very unscientific subjective tests I've favored fixed point) than with things like fades.  

Best regards,
Steve Berson

jfrigo

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1029
Re: 24 or 32fp in mastering?
« Reply #5 on: May 21, 2004, 05:25:21 pm »

OatBran wrote on Thu, 20 May 2004 19:16

It is the use of the fixed or floating in processing that I am particularly refering to.

I am not sure if PT is fixed or floating.  Might need to check on that.  Though I am sure Sonic is fixed.  I know Steinberg products are all floating and tracks that way as well (unless told to do otherwise).  I know Sonic Foundry/Sony products are fixed.  I am pretty sure Logic is floating though I am not positive on that one either.



Pro Tools HD TDM systems are 48 bit fixed point, while Pro Tools LE systems are 32 bit float. You can still have a plugin that calculates at 64 bit float but is fed and returned at 32. Sonic is also 48 bit fixed point. You may notice that Sonic and Pro Tools process on cards tend and are fixed point (not that all hardware has to be fixed point) while programs that do processing done by the CPU are typically floating point.

As a delivery format, 24 and 32 are essentially equal, though 24 is likely to be more widely compatible. Storing at 32 is mostly for CPU processing programs that are storing it as an intermediate file between processing steps. As somebody already alluded, 32 bit float is 24 bits of audio data with an 8 bit mantissa.

All other things being equal, 32 bit float processing will beat 24 bit fixed processing, though it is widely held that 48 bit fixed wil beat 32 float (and I agree). Dr. James A. Moorer wrote a brief paper on the subject if you want to do a search for some reading. 64 bit float gets floating point back in the game. While 64 certainly has a theoretical measurement advantage with the moveable zero, buy the time you reach 48bit fixed, the extra room in 64 float is pretty likely to go unnoticed. Both 48 and 64 are excellent for processing.

Some processors like the Weiss EQ and Studer digital consoles utilise sharc chips using 40 bit floating point precision. Note that some other sharc implimentations process at 32 bit float. 40 bit float probably is indistinguishable from 48 bit fixed or 64 bit float in general practice.

The really important factor is how the designers and programmers make their hardware and software. A crappy 64 bit plugin is going to sound, well, crappy. Excellent 32 bit processing can sound excellent. Just about the only thing that is avoided these days is 24 bit fixed point. With the industry using 24 bit files extensively, you really want a little bit more processing headroom for them (as a general rule you should process at a greater resolution than your source file), and it's so simple to use greater than 24 bit fixed point precision these days. There's no real reason to limit yourself. I also would imagine that 32 bit float processing will eventually give way to 64 bit float in the CPU, or  in hardware, 40 float, 48 fixed or 64 float. Now if only we had a 32 bit transmission channel...
Logged
Pages: [1]   Go Up