Re: [Audacity-devel] Amplify is very noisy

[Martyn S. <mar...@go...>]

Hi Leland

OK, so I know you claim to have no knowledge of signal processing, but 
this is above the level of 'trivial', so I forgive you!

Leland wrote:
> 
> --On June 2, 2008 1:48:19 AM +0100 Martyn Shaw 
> <mar...@go...> wrote:
> 
>> Maybe it isn't an issue.  What are you seeing?  I see 'silence' as a
>> flat line of zeros here.  Other places may see it as +-1 LSB (which I
>> would argue is probably more valid), but I don't know if others have,
>> or have coded that.  http://en.wikipedia.org/wiki/Dither seems a
>> reasonable reference if dither is the answer.
>>
> I think may we have a rounding difference.

It looks like it, but it's by design (see below).

> First time through, 
> "Amplification" is 0.0 and "New Peak Amplitude" comes up as -Infinity. 
> Tracing shows that the Init() method shows the single silent track's min 
> and max values are both zero.

This how it should be.  A silent track (exactly zero) has a dB level 
of -Infinity (as the log of zero is -infinity).  So any gain added 
would leave the level the same.

> Second time through, min equals -0.000274658203 and max equals 
> 0.000274658203, 

Now how did those values get there?  I deduce that your track is in 
16-bit PCM mode before generating silence.  Also that you have the 
default 'shaped' pref set in the Quality pref for dither.

What happens is that the Amplify processing is done in 32-bit mode, 
then converted back to 16-bit (...cut 5 lines of how this happens...). 
  Rather than truncate the data (which results in distortion), a more 
devious algorithm (dither) is used which turns the distortion into 
noise, less audibly objectionable.  The 'shaped' bit refers to the 
fact that this noise is made to have it's energy at higher frequencies 
where it is less objectionable still.  OK, in this case it looks bad, 
but on any real-world signal it's a good idea.  You don't want to get 
into the maths of this!

> giving an "Amplification" of 50 and a "New Peak Amplitude" 
> of -21.2.

The max amplification you can do at a time is 50dB.  Given the values 
that you indicate (0.000274) that will leave just that sized signal, 
so it's right, the signal is 71.2dB down on the maximum after the 
dither.  (20*log10(0.000274) = -71.2dB)

> I'm not going to look much further just now, but I reckon this might should 
> go on the "Things to look into list".  (I don't know where that is these 
> days.)

This isn't a 'problem', it's by design.  It shouldn't go on any list 
of problems.

HTH
Martyn

> Also, for the sake of novices, wouldn't a "No limit" checkbox be better 
> than "-Infinity" or would that mean 2 different things?
> 
> Leland
> 
> 
> 
> 
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Audacity-devel mailing list
> Aud...@li...
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>