Re: [Audacity-devel] Analysing ultrasound recordings with Audacity

[Thor A. <ta...@to...>]

On Sun, May 01, 2011 at 11:54:26PM +0100, Martyn Shaw wrote:
> On 30/04/2011 00:49, Martyn Shaw wrote:
> > On 27/04/2011 17:35, Thor Andreassen wrote:

[...]

> > > o The 'Maximum frequency' in 'Preferences->Spectrograms' shouldn't
> > >   be capped at 100 KHz, attached patch sets it at 10 MHz. Maybe the
> > >   cap should be removed?
> >
> > Your patch works fine for me with your file. I'd have to look harder
> > into whether we should have some cap on it.
> 
> There has to be some cap, due to the size of integers (see below).

Ok.

> It needs to be set with reference to the max size of the fft window as
> well, 

Maybe I'm being dense, but could you explain how these are related?

> and possibly something else we have not thought of, so we should be
> careful.

Yes.
 
> > > o Add more FFT window sizes.

[...]

> > > My patch causes a segfault when zooming out on the frequency axis
> > > of the spectrogram, this is also the case when enabling
> > > EXPERIMENTAL_FIND_NOTES. Not sure how to fix.

[...]
 
> We have  an overflow in the arithmetic in lines 1809 & 1810 with these 
> bigger windowSize and minFreq/maxFreq values, which are both int.
> Multiplying them together can cause an overflow with the larger
> values, say windowSize = 32768 and min/maxFreq > 131072Hz.  Locally I 
> have fixed this with:
> TrackArtist.cpp lines 1809-1810
>     int minSamples = int ((double)minFreq * (double)windowSize / rate 
> + 0.5);   // units are fft bins
>     int maxSamples = int ((double)maxFreq * (double)windowSize / rate 
> + 0.5);
> 
> Does that work for you?  

That's the problem, your solution fixes it.

> Is it the best way of doing it?

One cast should suffice, and float should be accurate enough:

  int maxSamples = int ((float) maxFreq * windowSize / rate + 0.5);

> I think the the result will not exceed an int since maxFreq/rate is
> probably 1/2 at most.

I agree.

[...]

> We are in a state of 'feature freeze' at the moment, trying to get to 
> that elusive 'Stable Release', and so one big patch is unlikely to get 
> accepted into Audacity at the moment.  Why don't you break this down 
> into different elements that could be applied in series, each one 
> checked for problems as we go along. Start with the renaming patch.

Ok.

[...]

-- 
best regards
Thor Andreassen