Re: [Audacity-devel] Clipping indicators anyone?

[Martyn S. <Mar...@ao...>]

Leland wrote:
> Quoting Martyn Shaw <Mar...@ao...>:
> 

>> Also, when I used the 'Find Clipping...' menu item on a rather clipped
>> 20s piece it generated rather a lot of labels.  I tried the label editor
>> which took 3m 32s to come up (I thought it had hung) and there were
>> 11051 entries.  I'm not sure this is reasonable behaviour, or why it
>> took so long to appear.
>>
> That's what the parameters are for, though I haven't a clue as how to describe
> them as such.

I realised that.  I can't think of better text for the dialog.

> The first parameter is the detection threshold in samples.  The plugin will
> ignore any runs of clipped samples that are less that this value.  In other
> words, there has to be this number contiguous samples that are clipped before
> the plugin will consider it.  The higher the number, the less likely a clipped
> region will be recognized and the few labels that will be generated.
> 
> The second parameter is sort of the same thing, but it is the number of
> UNCLIPPED contiguous samples that the plugin use to determine the end of a
> clipped region.  The higher the number here, the longer a region will be and
> the fewer labels that will be generated.

That's how I would have done it also.  The downside is that a 1s, 1kHz 
signal that is clipped results in nearly 2000 labels and a significant 
wait for the label editor dialog to appear.  Perhaps there should be a 
maximum number of labels parameter?

>> Also, it is only giving me an indication of clipping on negative peaks
>> when I record from a CD - playing it in WMP and recording the wave out.
>>
> This one I haven't figured out yet.  I noticed it too, but only on some things. 
> I have some MP3s that show positive and negative clipping.  But, recording only
> seems to show negative ones.  Any ideas would be appreciated. :-)

Once I had recorded a clipped-but-not-showing-positive-ones piece, I 
used 'Normalise' and put a break point in after it has found the max and 
min values (around li 252).  I find the min at -1.0 but the max at 
(float) 0.99996948, which is about 1/32768 off 1.  It's them clipping, 
rounding and conversion of integer to float values that's to blame, I'm 
quite sure, whether it's in hardwear or softwear.  We can't do much 
about either, in terms of where our audio arrives from.

It definitly varies with the input source as with my mike the max AND 
min are at +-0.99996948, almost exactly 1 (16 bit) sampling unit off the 
max/min.

>>   When I try my mike it does not give any indication of clipping, even
>> when there is severe distortion.  I think this is a function of the
>> hardware and the same thing that Gale and I were discussing a few weeks
>> ago with the meter not giving a clip indication.  My proposal to fix all
>> these problems is to detect x >= 0.999, rather that x>= 1.0, or
>> something like that.  What do you think?  The '0.999' could perhaps be
>> 1-(one sample at the current sampling depth)?

I said 'current sampling depth' here meaning the (voltage) resolution, 
ie 8 bit, 16 bit, 24 bit etc.  We could easily fix it for 16 bit devices 
but for other it may be more of a problem.

> How about making it a parameter in the "Find Clipping" plugin?

Hmm.  Would that then affect 'Show Clipping'?  If so, how would somebody 
find where to set it?  And it would take somebody better than us to 
figure out the best value to use!  This does not sound like the right 
thing to do.  I think a 'hard coded' 'just less than 1' is a better way 
to go, but see above.

I think that this is an INSERT_EFFECT, as it inserts labels.  Then it 
will get a 'Cancel' button.  What do you think?

TTFN
Martyn