Recording 32bit(float) vs. 24bit

Discussion in 'Digital Audio & Recording' started by oliverbuhr, Nov 1, 2002.

?

With how many bits do you record generally?

  1. 16bit

    6 vote(s)
    50.0%
  2. 24bit

    4 vote(s)
    33.3%
  3. 32bit / floating point

    2 vote(s)
    16.7%
  4. don\'t know

    0 vote(s)
    0.0%
  5. dosen\'t mind

    0 vote(s)
    0.0%
  1. oliverbuhr

    oliverbuhr New Member

    Messages:
    14
    Likes Received:
    0
    Trophy Points:
    0
    Hi,

    inspired by the thread about 16/24bit recording I was wondering if 32bit recording would increase quality, not just size.
    I'm using Cubase VST32, but i never used the option called "true tape" which seems to be only a sort of compressor.

    And I've never seen (or have I overseen?) a 32bit capable recordingcard until now.

    Does anybody of you use 32bit? If yes, for what reason?
    Last edited: Nov 1, 2002
  2. malsony

    malsony New Member

    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    0
    i use 24bit.
    it's ok, 'cos the difference could seldom be told
  3. BrothersGrim

    BrothersGrim New Member

    Messages:
    10
    Likes Received:
    0
    Trophy Points:
    0
    I agree with Malsony, the human ear can only hear so much. anything above 16 bit is pretty much a really HUGE file and you can't hear the difference... You hear about 8bit, 16bit, 24bit blah, blah but does anyone know what that really means? I do....:)
  4. jon-paul

    jon-paul New Member

    Messages:
    206
    Likes Received:
    0
    Trophy Points:
    0
    Well guy this is the situation. you only use 32 bit or 24 bit if you are going to be editing your recording alot and i mean alot. everytime you edit your recording by adding effects or cutting or adding eq bla bla bla you lose some of the quality. if you have recorded a 64 piece orchestra and each one of them needs editing then you could lose a bit of quality and so by recording in 32 bit this gives your editng monouvering space and still means that you end up with a recording that is at CD quality.

    I only know this because i am doing a Degree in Music Production in England. if you want to know more Email me directly

    Hope this has cleared up the conffusion about 32 and 24 bit recordings

    cheers jp
  5. acieed

    acieed New Member

    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    0
    IMHO, I believe you can hear the difference better when you're recording stuff such as live drums.
  6. Graeme

    Graeme New Member

    Messages:
    980
    Likes Received:
    0
    Trophy Points:
    0


    Well - almost right. If you are going to be doing a lot of post processing on a file, then 32 bit float is the format of choice, as it's scaleable and (for all practical purposes) you can't exceed the clipping point. When re-sampling down (as would normally be required to make a CD) the rounding errors, which are present in both 16 and 24 bit systems do not occur.

    The downside of all this is very large files, so you have to make a judgement call on the need to actually do this.

    To pick up on oliverbuhr's question - you'll never see 32 bit hardware, since it's potential dynamic range far exceeds the physical possibility of construction. People who want/need to use 32bit float files upsample them in softwae (eg Adobe Audtion) before starting work.



    I don't know what they teach you guys these days, but someone doing a degree in music production really should know that the processes you are referring to are not called 'editing' - that's something else entirely.

    I know this as someone who learned the business through practical application, not academia.
    Last edited: Apr 6, 2004
  7. Graeme

    Graeme New Member

    Messages:
    980
    Likes Received:
    0
    Trophy Points:
    0
    I should have also said, if you are truly only editing - ie cutting things up and re-positioning them - then it matters not one iota what format the track is in, since this process does not affect the quality, it will remain exactly the same.
  8. pwrie

    pwrie New Member

    Messages:
    2
    Likes Received:
    0
    Trophy Points:
    0
    Perhaps some insight can be gained looking at the math. With A/D convertors, By convention, full scale is taken as one (1.0). the Most Significant Bit, is set at ½ of full scale or 0.5 , the next most significant bit, is ¼, and so on. The math representative for this series is 1/ (2)^n for n = 1,2,3,.. infinity (using nomenclature from openoffice or libreoffice spread sheets, and 1/(Power(2,n)) for n = 1, 2, 3, … infinity - (using MS Excel spread sheet nomenclature)

    The full scale representation is actually per convention, or close to 1 minus the Least Significant (LSB). For a 16 bit sample this is 1/65,536 (0.000015259 in decimal) for 24 bits 1/16,777,216 (0.00000005960 decimal). The error of the measurement, is the just less than the Least Significant bit (the “truncated bits out to infinity, if the missing bit were added (summed) together all of these missing bits will fall just short under the least significant bit) value/weighting.

    Basically the quantization errors or (noise) of a 16 bit samples is just slightly less than 0.000015259
    the quantization error or (noise) of a 24 bit sample is just slightly less than 0.000000005960

    32 bit floating, has a “fixed” 24 bit (mantissa). There is also a 8 bit exponent which the best referenced at Wikipedia – look at “powers of two” therein. With a wide (large) dynamic range, the power can be adjust to get the most accuracy in the mantissa, the most significant bits are not zero.

    If filtering (formally call Digital Signal Processing (DSP)) the difference can show up, with one or more multiple multiplication/Division involved, and or additions/subtraction. The perhaps extreme case –take two fix point numbers in “filtering”.

    For example and obviously not an excellent example - multiply, the LSB of 16 bit (fixed) number against the LSB of a 24 bit (fix) number, using above: 0.000015259 times 0.00000005960 = 0.000000000000909 the results would rounded to zero on either the 16 fix for 24 bit fix representation. Using 32 bit (24 bit mantissa) floating point numbers representation for these two numbers the result will not be “rounded” to zero. If this results is then used to multiple a large number, the difference between fix and floating point is noticeable, one is zero, the other is no zero nor close to zero.

    If a software equalizer is used; that the place I would check for the difference in quality; and it obviously depends upon how the software is written, not just on the fix or floating point representation of the samples.

    Note, the digitized sample is NOT a floating point number, it in this case 16, or 24 bits. The significance, “if any” is in the mathematical operations performed on the sample(s) and the results and significance of the error in the results of the mathematical operations - after the sample is taken, i.e., equalizer, data compression, “noise reduction techniques”, smoothing or many other “special effects”.

    It might be informative use Wikipedia search term “ HDTV sound track format “ -(index - high resolution)

    Basically to go further, I suggest research Analog to Digital Converters and Digital Signal Processing. .
  9. posik

    posik New Member

    Messages:
    10
    Likes Received:
    0
    Trophy Points:
    1

Share This Page