| Author
|
poll. preferred bitrate opinions
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Aug 10, 2007 20:39
|
@sy000321
Both 24 bit fixed and 32 bit float is actually 24 bit precision, so one doesn't have more precision than the other as long as the signal is fully scaled.
As for the processing it's done in 32 float regardless of what format you use when recording.
@Colin
Of course...I was referring to recording down internally....forgot that people still use external hardware or real instruments 
Still as long you don't clip the converter you can add or subtract gain when in the digital domain and still have a full resolution and not have to worry about clipped signals if you record to 32 bit float.
  (``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
piXan
IsraTrance Full Member
Started Topics :
107
Posts :
807
Posted : Aug 10, 2007 21:42
|
ok nice info
gonnay try 32 bitfloat on cubase setup , as most of the recording is actually rendering or mixdown , from virtual instruments, and record of course at 24 bit for vocal guitar etc.
by the way and off topic..
whats the difference between rendering, mixdown and bouncing????? |
|
|
|
sy000321
IsraTrance Junior Member
Started Topics :
46
Posts :
1142
Posted : Aug 10, 2007 22:09
|
Quote:
|
On 2007-08-10 20:39, Spindrift wrote:
@sy000321
Both 24 bit fixed and 32 bit float is actually 24 bit precision, so one doesn't have more precision than the other as long as the signal is fully scaled.
|
|
erm... can you explain how that could be?
roll a joint or STFU :) |
|
|
|
sy000321
IsraTrance Junior Member
Started Topics :
46
Posts :
1142
Posted : Aug 10, 2007 22:43
|
32 bits floats (as defined by IEEE, doc 754  ) are stored using 3 operands:
1 bit for the signal (+/-)
8 bits for the exponent
23 bits for fraction
in this way, extremely big numbers and very small ones can be stored, adding precision 24 bits fixed point cant achieve
these small nuances are lost when converting to 24 bits, but it makes a difference while you're working with lots of sound sources, less rounding is made while in 32 bits
conclusion, 32 bits floats wins in every possible way
roll a joint or STFU :) |
|
|
|
Spindrift
Spindrift
Started Topics :
33
Posts :
1560
Posted : Aug 11, 2007 02:47
|
The actual resolution is contained in the 24 bit mantissa that you mentioned.
There you have your integer values between 0 and 16,777,216 that is the number of "steps" in which you can measure the waveform.
That number of steps is the same in 32 bit float and 24 bit fixed.
With 32 bit float you also have an 8 bit exponent, which is a decimal value that is used to scale the values in the mantissa.
The exponent is not used to make more possible steps, i.e. add more resolution, but to adjust the level of the information in the mantissa.
Even if the range is greater it's still the same resolution.
Say for example you have a measuring tape with millimetre precision and are measuring an object to make a scaled replica.
The measurements will not get better precision by multiplying or dividing them and will remain the same no matter how you scale it even if the range has changed.
(``·.¸(``·.¸(``·.¸¸.·`´)¸.·`´)¸.·`´)
« .....www.ResonantEarth.com..... »
(¸.·`´(¸.·`´(¸.·`´``·.¸)``·.¸)``·.¸)
http://www.myspace.com/spindriftsounds
http://www.myspace.com/resonantearth |
|
|
|
sy000321
IsraTrance Junior Member
Started Topics :
46
Posts :
1142
Posted : Aug 11, 2007 04:28
|
Quote:
|
On 2007-08-11 02:47, Spindrift wrote:
Say for example you have a measuring tape with millimetre precision and are measuring an object to make a scaled replica.
The measurements will not get better precision by multiplying or dividing them and will remain the same no matter how you scale it even if the range has changed.
|
|
actually, when when divide an integer number you often loose precision.
in floating point you may also loose precision, but it's a smaller error.
floating point is great for calculating very small numbers, imagine a reverb for instance where loads of tiny echoes fading away.
in fixed point, as the volume fades to 0, it starts to loose definition as it disappears
in a 32 bit float when scaling down you change the exponent, so you retain 100% accuracy
this nature of this subjects is complicated being computers totally abstract representations of nature, like if dbs and floats werent already difficult concepts to graps, its very difficult to calculate best case and worst case errors, but in the best case 24dbs is at least as good as 32float (for the same headroom)
using natural logarithms in fixed point can obtain the same precision as the discussed floats, but has a highly expensive performance hit to use for calculations so it's not used in real time applications (i believe).
mind you, precision is only kept while in 32 float.
im tired now
roll a joint or STFU :) |
|
|
|
Mike A
Subra
Started Topics :
185
Posts :
3954
Posted : Aug 11, 2007 04:33
|
Just to fix a small error - it's bit depth, not bit rate.
Bitrate is how much bits go in a second in mp3 files, for example. Nothing to do (almost) with 16 or 24 or 32 bits.
|
|
|
|
sy000321
IsraTrance Junior Member
Started Topics :
46
Posts :
1142
Posted : Aug 11, 2007 04:56
|
|
everyone overlooked that... hhehe
roll a joint or STFU :) |
|
|
|
sy000321
IsraTrance Junior Member
Started Topics :
46
Posts :
1142
Posted : Aug 11, 2007 05:00
|
Quote:
|
On 2007-08-10 21:42, piXan wrote:
whats the difference between rendering, mixdown and bouncing?????
|
|
rendering can be bouncing or mixdown, mixdown is usually the whole mix and boucing just one part/instrument/mixer channel.
some engineers will call mix down the process of mixing a whole track.
roll a joint or STFU :) |
|
|
|
Upavas
Upavas
Started Topics :
150
Posts :
3315
Posted : Aug 14, 2007 04:28
|
|
