Close

Page 11 of 12 FirstFirst ... 9101112 LastLast
Results 101 to 110 of 116
  1. #101
    Join Date
    May 2004
    Location
    St. Petersburg
    Posts
    3,842

    Default Re: Overs... working correctly?

    Im never going to eat jelly again you can be sure of that....



    mostly because Ive hated the texture of gelatin my whole life haha
    Lovingly signed,
    Robert Randolph

  2. #102

    Talking Re: Overs... working correctly?

    I too would be interested in reading your Papers and findings Canipus. If you could provide a link i would love to check it out!
    Carey A. Langille
    Producer/Engineer
    Ocean Sound Productions
    www.oceansoundstudio.com
    Creativity at the Speed of Sound

  3. #103

    Default Re: Overs... working correctly?

    Just using a DX plugin pretty much throws you into floating point math... many VST and ASIO implementations are forced into floating point as part of the design... most all of the industry seems to think you need floating point to add 1.5 to 1.5 to get 3... integer would give a result of 2 since there would be roundoff of 1.5 to 1... oh well... such is the hype that floats around.

    I really don't know who uses what... but I do know there is a big difference in the results... at least to my ears... and that's all that matters when I choose how I would like to do my audio.

    I have tripped over many instances of floating point math in the areas where I use it... like the timeline... where the results are demonstratably different from one pass throiugh the code to the next pass through the same code. Seeing this as a developer keeps me very leary about using it in my code.

    Floating point is extremely sensitive and requires an incredible amount of attention to each use of each FP instruction and when its used and in what order its used and what the rest of the machine register conditions are when its used... that to get it right becomes an enormously complex task... and from what I see... most of the software out there uses standard libary calls and patches together pre-written functions and modules and C++ class objects without ever being a part of the code development of those modules... the results can be pretty messy and very innacurate... and extremely inefficient and slow to boot.

    The use of integer math by some of the top high-end audio manufacturers was never really common knowledge a few years back... it seemed to not be mentioned by these companies much... perhaps becuase it was not in fashion with the accepted audio standards... and we wouldn't want to upset belief in the industry standards now would we.

    And Yura... I hate to disagree... but testing max limits of things has no real bearing on their use or efficiency under normal working conditions. in my opinion..

    I mean... just because a beautifu red Ferrari does not stand up to driving it through a brick wall, hardly negates its value and enjoyment quota when driven under normal driving conditions down a freeway.

    Worrying about useless headroom of a signal in FP math has no bearing on the quality or efficiency of using the math to do a real mix... if you feel the need to have input signals 60db above clipping... I hardly imagine that quality sound and music is the primary output focus.

    I want the most efficient and most acurate results of my signal math all the way through the chain to the final mix... period... regardless of the fact that some theoretical unusable headroom limit is higher using slower and more inacurate math models.

    Then again... this is just my opinion... and obviously not the consensus of the industry in general... so what else is new.

    Bob L

  4. #104

    Default Re: Overs... working correctly?

    Quote Originally Posted by Bob L
    Just using a DX plugin pretty much throws you into floating point math... many VST and ASIO implementations are forced into floating point as part of the design... most all of the industry seems to think you need floating point to add 1.5 to 1.5 to get 3... integer would give a result of 2 since there would be roundoff of 1.5 to 1... oh well... such is the hype that floats around.

    I really don't know who uses what... but I do know there is a big difference in the results... at least to my ears... and that's all that matters when I choose how I would like to do my audio.

    I have tripped over many instances of floating point math in the areas where I use it... like the timeline... where the results are demonstratably different from one pass throiugh the code to the next pass through the same code. Seeing this as a developer keeps me very leary about using it in my code.

    Floating point is extremely sensitive and requires an incredible amount of attention to each use of each FP instruction and when its used and in what order its used and what the rest of the machine register conditions are when its used... that to get it right becomes an enormously complex task... and from what I see... most of the software out there uses standard libary calls and patches together pre-written functions and modules and C++ class objects without ever being a part of the code development of those modules... the results can be pretty messy and very innacurate... and extremely inefficient and slow to boot.

    The use of integer math by some of the top high-end audio manufacturers was never really common knowledge a few years back... it seemed to not be mentioned by these companies much... perhaps becuase it was not in fashion with the accepted audio standards... and we wouldn't want to upset belief in the industry standards now would we.

    Bob L
    Bob

    I tend to agree with a lot of what you say. Many filter algorithms and phase
    accumulators are not very stable with 32bit float. However, re. plugins in general could part of the problem be simply the common use of a 32 bit engine rather than floating point versus integer? I've noticed many of the plugins with 64 bit processing are very useable and as high quality as anything my 58 year old ears are able to tolerate. I have noticed that if you use more than 3 or 4 common or garden 32 bit plugins on a channel the audio loses its edge. With 64 bit there seems to be more leeway. On the other hand maybe the real issue is whether one adopts single or double precision. It would make a lot of sense if the quality improvements were centered around double precision maths but again I have no idea if the manufacturers working with 64 bit float are using single or double precision. Many of them just don't bother to declare what they are doing and I had an incredible situation a couple of days ago where some of them weren't even certain. This following is absolutely true.
    I was being interviewed (by phone) for a VP position to head up the business management for a new music s/w product. It turned into quite a conference with their CEO/President (a woman who actually seemed to know what she was doing), VP Strategic Planning Engineering, and some of the project management group. When we got onto discussing the mix engine I asked the question whether the engine was based on integer math of floating point. Would you believe instead of an answer I got 5 seconds silence followed by some whispering and mumbling and then some throat clearing until I finally got the answer "floating point" I then had to ask the follow up question, "how many bits" and we went through a repeat performance of mumbling and throat clearing.
    It's makes you smile doesn't it...Wonder will never cease!

  5. #105
    Join Date
    May 2004
    Location
    St. Petersburg
    Posts
    3,842

    Default Re: Overs... working correctly?

    There is a such thing as single precision 64-bit floating point?

    Hrm... new one on me then lol Single precision is 32-bits, double precision is 64-bits.

    More or less in terms of audio 32-bit integer offers the same quality as 64-bit floating as it is used in 99% of applications today. THOUGH double precision floating can be more accurate, it is much harder to use.. and im not aware of anyone who is using it properly.

    I think it can matter quite greatly when you start getting lots of tracks with even just fader and pan... toss in plugins and.. well.

    Since Ive gone 100% saw native format Ive noticed a huge difference in the type of sound Im getting. It's quite interesting... I even made myself a plugin that simply did a large gain change down then up in the floating point realm (high resolution) and patched it on every track of a 30 something track project. IT sounded totally different.. and indeed an anlyzation of the buildmix showed "audible" differences (above -70db and peaked difs up to -50db). But all in all I dont really care that much anymore... I got saw, it works great in everyway I need it too.. so that's end of story.
    Last edited by AudioAstronomer; 03-02-2005 at 02:40 PM.
    Lovingly signed,
    Robert Randolph

  6. #106

    Default Re: Overs... working correctly?

    Quote Originally Posted by AudioAstronomer
    There is a such thing as single precision 64-bit floating point?

    Hrm... new one on me then lol
    Sorry Robert, my head is too wrapped up in hardware. One of the ways the semi industry tried architecting 64 bit DSP back in the time when heat was a problem on the die was to fiddle around with two 32 bit chips to get 64 bits and by some jiggery pokery get a double precision computation. However the individual chips were single precision processing and the computations were pseudo double precision. Ever since i have had it my head that you can get single precision 64 bit (instead of following the IEEE definition) because in REALITY that's what this was.
    My bad!

  7. #107

    Default Re: Overs... working correctly?

    Quote Originally Posted by AudioAstronomer
    There is a such thing as single precision 64-bit floating point?

    ... I even made myself a plugin that simply did a large gain change down then up in the floating point realm (high resolution) and patched it on every track of a 30 something track project. IT sounded totally different.. and indeed an anlyzation of the buildmix showed "audible" differences (above -70db and peaked difs up to -50db). But all in all I dont really care that much anymore... I got saw, it works great in everyway I need it too.. so that's end of story.
    Robert you should care! That is valuable work and with further substantive testing and collation is worth documenting and publishing in the AES journal.

  8. #108

    Default Re: Overs... working correctly?

    Quote Originally Posted by canipus
    When you compare DAWS you need to keep in mind the difference between processors that utilize floating point scaling rather than fixed integer. Generally speaking any engine that uses 32 bit floating point unbounded single precision, will give you virtually unlimited headroom against clipping audio and more dynamic range than you could ever possibly need. The whole ideology of floating point is that the scale adjusts up or down in order to cover whatever range is required in the calculation. That isn't to say integer math can't do the same thing. It can, you just need more bits.
    Ok.
    What for.... What for, as for us, being literate men, to do consider that those comparisions about matter of dynamic range between the two ways of processing...to consider them as being usefull?
    when for you and me - it is quite understandable that in both cases the dynamic range is sufficient to kill the elephant and to deafen the dolphin?
    What are advantages of the endless dyn. range that is even "more dynamic range than you could ever possibly need" as you pointed by yourself?

    That is something of the kind of possible researches on the subject of linear characteristic of the post DA filter dependency on the temperature of surrounding environment, being deviated within 0.01% and measured to be related with a human's ear perception, having inaccuracy MORE then 0.5db= 5% (rough-average)!!! So often measure an ant with a 1 kilometer-compasses.(what is the temperature coefficient of the human ear?)

    I just did that test to reinsure about sufficientness of internal SAW dyn range to terms of AES and I have made sure I was not correct when I distrusted about it before due to some earlyer thoughts. Of course, Bob must be LOLled of such a thoughts and actions, but if to cite to AES figures (and a sane sense) it would be very logical to see such a gesign of signal path looks solid-reasonable. In terms of range sufficiency, and engine reliability.

    Quote Originally Posted by canipus
    This link may help give more background on the subject
    http://www.ngwave.com/comparisons/so...comparison.php
    Anyway, thanks for a link. No doubt, interesting to see.
    But you know, again, there are lot of another links in favour of integer math big advantages...
    Last edited by Yura; 03-02-2005 at 03:46 PM.
    George Oran'sky

  9. #109

    Default Re: Overs... working correctly?

    Quote Originally Posted by Yura
    Ok.

    Anyway, thanks for a link. No doubt, interesting to see.
    But you know, again, there are lot of another links in favour of integer math big advantages...
    yura I sent you the link to illustrate some of the tests that can be done to illustrate the meaning and advantages of headroom relative to math processing and not specifically to eulogise the benefits of floating point over integer or vice versa

  10. #110

    Default Re: Overs... working correctly?

    Quote Originally Posted by canipus
    yura I sent you the link to illustrate some of the tests that can be done to illustrate the meaning and advantages of headroom relative to math processing and not specifically to eulogise the benefits of floating point over integer or vice versa
    Thank! Very!

    Do ya know!
    what for... seriously..., my real goal is to see how this meter will show 5.000 (thread)
    Has it any limit?
    George Oran'sky

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •