On Sun, 12 Aug 2012 18:02:59 +0200 Tom Gundersen <teg@xxxxxxx> wrote: > On Sun, Aug 12, 2012 at 5:27 PM, Fons Adriaensen <fons@xxxxxxxxxxxxxx> wrote: > > On Sun, Aug 12, 2012 at 04:00:47PM +0200, Tom Gundersen wrote: > > > >> You have showed that it is unnecessary in one particular (very simple) > >> case. However, you have not showed that it is unnecessary in all > >> cases, so this is not really relevant (had we been talking about a > >> human doing this, you'd have a point of course). > > > > I suspect that mathematical thinking is not your thing - no > > problem. For otherwise it would be clear that the 'simple' > > example I provided covers the general case. > > > > Let me try again. I write a PA-aware sound app X that > > > > * always sets its volume to 0 dB (max). > > * always outputs silence (zero valued samples). > > > > As soon as that app runs, PA will set the master gain to > > 0 dB and use software scaling on all other apps. Now there > > are two possibilities: > > > > * Either everything is OK (it will be), and we have shown > > that you can always leave the master gain at 0 dB, > > > > * or everything is not OK, and we have shown that PA > > fails. > > Your argument was clear (I'm ok with maths :) ). I was just pointing > out that the case where the hardware can always be set to 0dB is not > the interesting one. > > You clearly want to avoid setting the hardware to >0dB if possible, Correct me if I'm wrong but I don't think that's possible, because dB is normalized to max power (in watts = intensity). > but sometimes it is not possible (because you can't hear anything ;) ) > so you have to have an algorithm to do it in the best way. Imagine you > have two streams (A) which needs no software nor hardware gain, had it > been played alone, and (B) which forces the hardware gain to be >0dB > (and (A) to be scaled down in sw). If (B) goes away you clearly want > to set the hw volume back to 0 and (A) to stop being scaled in sw. > > The second case, where the total gain should be <0dB, I would have > thought intuitively that doing this purely in software (especially on > very faint signals) would be less ideal than doing it in hw (you'd be > throwing away the resolution, wouldn't you?), but I'll admit that I > don't have the experience to talk about that with any authority. > > -t -- Leonid Isaev GnuPG key: 0x164B5A6D Fingerprint: C0DF 20D0 C075 C3F1 E1BE 775A A7AE F6CB 164B 5A6D
Attachment:
signature.asc
Description: PGP signature