Subject: RE: [linux-audio-dev] normalize: destructive or non-destructive ?
From: MOULET Xavier FTRD/DMR/ISS (xavier.moulet_AT_rd.francetelecom.fr)
Date: Fri Nov 03 2000 - 17:58:33 EET
maybe I need a little explanation here ...
what do you exactly mean by 'normalizing' ?
I thought that it was "amplifying up to the fact that one sample reaches the
max"
So only one number has to be determined : the amp. gain.
If you want to normalize with an envelope, you will define a compressor, I
think, and here you just have to store a limited number of points, because
your envelope can be defined with wuch less points, interpolated for each
sample (of course not necesserally linearly, can be cubic interpolation). An
envelope has necesserally a lower frequency than the original signal, hence
a lower sampling freq. And if you can infer the envelope from the few
points, you can undo easily by recalculating the envelope on the fly.
If you want to have ONE amplifying gain for each sample, so that each sample
will be exactly amplified to 1 (normalization) , er ... I do not think that
the constant signal '1' will be what you intended to record.
Or do I (and possibly other) need more explanation ?
Maybe we should know what is acceptable to do ("i would not plan on
automating that ") to undo an operation. Because inverting the global gain
is what I found obvious .. I must be forgetting something.
well ,thanks a lot for your great work.
xavier
> -----Message d'origine-----
> De : Paul Barton-Davis [mailto:pbd_AT_Op.Net]
> Envoyé : vendredi 3 novembre 2000 15:36
> À : linux-audio-dev_AT_ginette.musique.umontreal.ca
> Objet : [linux-audio-dev] normalize: destructive or non-destructive ?
>
>
> in considering how to implement "normalize" for Ardour's editor, i am
> faced with a dilemma. if done non-destructively, this operation
> *necessarily* results in a complete duplication of all data currently
> making the sample stream for each track. it therefore seems to me that
> this operation should always be done destructively. if i do this to
> the recordings i'm working with even as test cases, we are talking
> about copying 1.2GB of data just for this operation. in the real
> world, we'd have, say, a 40 minute session occupying 12-18GB and would
> be required to copy all this data just to normalize. Insane.
>
> If necessary, it can be undone by knowing the scale factor used to
> normalize, and rescaling with its reciprocal, but i would not plan on
> automating that (it doesn't fit into the undo/redo/apply
> model i have).
>
> opinions ? am i missing something ? oh, and before anybody suggests
> it, its no help storing a gain curve for the whole thing. there has to
> be one element in the gain curve for every sample, and since
> sizeof(ARDOUR::Sample) == sizeof (ARDOUR::Gain), nothing is gained
> (pun not intended).
>
> --p
>
This archive was generated by hypermail 2b28 : Fri Nov 03 2000 - 18:47:57 EET