1. IMPORTANT:
    We launched a new online community and this space is now closed. This community will be available as a read-only resources until further notice.
    JOIN US HERE

Absynth Strange Rendering

Discussion in 'Technical Issues - KOMPLETE (Archive)' started by FAC, Mar 14, 2012.

  1. FAC

    FAC New Member

    Messages:
    4
    I've discovered this strange issue in the last demo version of NI Absynth.
    Is it a bug?

    Thanks you very much in advance,

    1) Open Absynth
    2) Select a blank patch (only one OSC)
    3) Hit record button
    4) Play C3
    5) Export the file in stereo
    6) Open the exported file in any wave editor
    7) Invert Left signal
    8) Sum Left and Right channels (mono)
    9) Normalize and you will get this horrible noisy sound!

    Screen shot 2012-03-01 at 18.40.02.png
     
  2. FAC

    FAC New Member

    Messages:
    4
    Is there any NI Absynth's user that could confirm this bug please?
     
  3. dreamkeeper

    dreamkeeper NI Product Owner

    Messages:
    492
    Why do you assume it's a bug? Can you hear anything without normalising?

    What you've done is boosting the difference between the two channels by a ridiculous amount. All digital signals have quantisation noise, but it's far far down in level (lower than -140dBFS with 24bit). By normalising you've brought that noise up to 0dBFS.

    "Doctor, it hurts to do this..." - "So don't do it then."
     
  4. FAC

    FAC New Member

    Messages:
    4
    No sorry! I've performed the same test with other virtual instruments and I've got nothing (as expected).

    For example in "wave editor" (a respectable editor):

    1) Generation of a sine (On two channels)
    2) Inversion of left channel
    3) Summing
    4) Normalization

    Result: You guess what? I've got nothing...

    So to resume according to you it's normal to get a difference between the two channels of a (mono) material (no panning, no reverb...)? Note that you don't have to put the normalization hard. A little amount of it produces this noise...

    Personnaly I find that Absynth is a great product but I have the feeling that it sounds harsh... Seriously, the issue I've presented you is a piste to investigate.

    Regarding doctors... I avoid them, in general all they do is to steal your money. :D
     
  5. dreamkeeper

    dreamkeeper NI Product Owner

    Messages:
    492
    Depending on how the signal is generated, there might be differences between channels. But as said, they shouldn't be greater than the quantisation error.

    I'm not sure what you mean by "a little amount" of normalisation. Either you normalise or you don't - anything inbetween would be just raising the gain by an arbitrary amount. Now it'd be interesting to know by how much (dB).

    Normalising by definition means raising the overall gain until the highest peak reaches 0dBFS. If those peaks would be down at -140dB, normalising would result in a boost by 140dB. I'll investigate though and let you know.

    And I agree about the doctors - and would add that they even steal your health. ;)
     
  6. FAC

    FAC New Member

    Messages:
    4
    Sorry I meant this:

    Screen shot 2012-03-17 at 15.46.11.png

    Thank you, I really appreciate that you'll investigate though and let me know the result. :)
     
  7. dreamkeeper

    dreamkeeper NI Product Owner

    Messages:
    492
    OK. First off, Absynth's recorder renders 16bit files, so one would expect a SNR of about 96dB. Doing your test, I'll get -90dB channel difference - which shouldn't be a problem. Btw: it's not a mono file. Even with Absynth's signal path set to mono, a stereo file will be rendered.

    When rendering to floating point in a host (Sonar), the noise is down at ca. -120dBFS - so even less of a problem. That the theoretical accuracy isn't reached, may be caused by the internal signal path in Absynth.

    It seems that you normalised to -24dBFS, This will boost any signal, no matter how low*, to peak at -24dBFS - which of course is audible. But when normalising in a practical situation, the peaks are given by the usable signal. And the latter will be (usually) far higher than the noise, so that the noise still would be inaudible.

    Bottom line: "Don't do it!" (normalise noise in isolation) ;)
     
    Last edited: Mar 17, 2012