1. Register NOW and become part of this fantastic knowledge base forum! This message will go away once you have registered.

The 32bit Float Advantage

Discussion in 'Recording' started by Giovanni_Speranza, Nov 15, 2005.

  1. Import a music track into a 32bit float engine audio software
    Put 2 gainer plugins on the track and lower the first gainer by 143dB
    boost the second gainer by 143 dB
    Listen, it sounds exactly the same.
    And now compare it with the same track phase inverted: silence.

    Now to see how it would sound in a 24 bit fixed environment, just bounce the -143dB track to a 24bit file, import it and apply 143dB gain on that file (Warning, lower the monitoring level!!!!!!!!!!) . It will sound totally crushed
    Simpler if you have a protools there, just try the gain tip and prepare yourself for the same crushed sound.
     
  2. iznogood

    iznogood Guest

    i will try it.... good idea :cool:

    btw..... i almost get nauseous looking at your avatar picture :) )
     
  3. hociman

    hociman Active Member

    flawed

    Your methodology is flawed. If you have a 24-bit audio signal, and lower the output of the channel with that signal by -143dB, then you are left with, effectively, a 1-bit output (not even, but I'll leave it to you to figure out the math)! 23 of your 24 bits will not be used. I am confident if you had some digital audio test/measurement equipment that you would see this.

    By importing that bounced track back into a DAW, and applying 143dB of gain, you are doing nothing more than making your 1-bit (effective) signal from the prior paragraph extremely loud. It will have no dynamics (didn't you say crushed?) because it is only, effectively, a 1-bit signal that you applied maximum gain to.

    The real way to do this test would be to do what you did in the 32-bit environment in the 24-bit environment. You may be thinking, "Well the 24-bit environment has 48-bits of precision between input and output. That's not fair." Well, it doesn't matter. A well-written and implemented algorithm performs equally well in either system.
     
  4. If i would do the same test with 24 bit floating point (instead of 32) the result would be exactly the same (obviously with 24 bit resolution instead of 32).
    The matter is floating vs non floating.
    What i'm trying to let you understand is that fixed point is flawed. Floating point gives you at any level the SAME RESOLUTION that you would have at full level.
     
  5. hociman

    hociman Active Member

    not so

    How many bits you have in your floating point math is irrelevant with your flawed test. Fixed point mathematics are not flawed, they are limited, and they are limited by the fact that they are fixed. Floating point operations are not superior to fixed point operations when the algorithms are done properly. Perhaps you are using poorly designed algorithms? Floating point operations have their drawbacks too, but that does not make them flawed.

    To claim that float is superior to fixed (or vice-versa) is like comparing apples and oranges. There is more than one variable - the number of bits and the operations. If you were claiming that 24-bit PCM was superior to 16-bit PCM (where we have one variable) you would have a valid statement.

    I have to question how much you really know about floating-point mathematics. Did you know that a 32-bit floating point word has a 24-bit mantissa (your 24-bit fixed point word), and a 7-bit exponent? The last bit, the sign bit, is 1 for a normalized mantissa and is not stored. The 7-bit exponent gives you a large range, but not the SAME RESOLUTION at a lower level. You can represent a number several different ways with scientific notation (what floating point is), but a number is a number. What matters is how many significant digits you have, and while floating point may have significant digits beyond the decimal point, floating and fixed point have the same maximum number of significant digiits.

    ;)
     
  6. Re: not so

    As we are not talking about plain audio reproduction, but about massive audio processing , it's a valid statement.
    Your thesis is exact, but if the early engineers would have considered the valve's technical properties the same way you are approaching the question, there would have never been an analog recording to date. The approach for audio processing is sonic quality. And my test proofs that 32bit floating point is an advantage for mixing audio.
     
  7. hociman

    hociman Active Member

    fixed mix

    Have you sat in front of a TDM system and done this test without bouncing? Since TDM is fixed, you should get the same results without bouncing, right? Put your file on a track, and lower its fader. Route that lowered output to another track, and raise its fader by the absolute value of the first fader.

    If it still sounds like garbage, let me know. Until you do this test this way, I cannot give your results a stamp of approval.
     
  8. Re: fixed mix

    it will work
    when i still used PT TDM i did a similar test:
    i layered the same song on many tracks and every track's volume was set to -60 dB. The resulting sound was almost noise.
     

Share This Page