Re: [Audacity-devel] Zooming in to Spectrum view - committed

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

Hiya
 
Ok, so at my second attempt (only my second attempt!  I must be  getting 
better!) I managed to make this work.  The news is good and  bad (but mostly good 
here, at least).  I tested with a 10s piece of audio,  with Audacity 
full-screen at a 1024x768 resolution.  The results are  unsurprising since I have 
swapped the screen resolution for the FFT size in the  cache:
 
Switching from a waveform view to a spectrum with an fft size of 4096 is  
about 10% slower (not good, more samples to calculate and cache in  
Spectrum::ComputeSpectrum).
Switching from a waveform view to a spectrum with an fft size of 256  (the 
default) is about 10% faster (good, less things to calculate in  Spectrum).
Resizing the spectrum window when it's 'large' is about 4-10 times faster,  
when it's 'small' is about 18-20 times faster (no calculations in Spectrum,  
only a bit more in TrackArtist).
Vertical zooming is 4-14 times faster, depending on window size, and now  
quite usable (I didn't think it really was before).
Horizontal scrolling is 1-1.1 times slower (not good, but not too bad  
either).
 
Overall I thought this was OK so I checked it in.  Hope you  agree.  Any 
further ideas?
 
TTFN
Martyn
 
In a message dated 21/01/2007 09:58:38 GMT Standard Time,  
do...@au... writes:
Martyn,

I'd have to think about the  exact details, but I totally agree with  
the general idea that the  spectral values should be cached, rather  
than the resized spectrogram  image.  That should also make vertical  
sizing of a spectrogram  much faster, too.

- Dominic


On Jan 20, 2007, at 2:44 PM,  Mar...@ao... wrote:
> Hi
>
> I implemented the zooming  in to the spectrum view but it is very  
> slow and does not take  advantage of the clever caching arrangement  
> in  WaveClip::GetSpectrogram.  I have come up with a plan to  
>  increase the speed and wondered if anybody has an opinion on it  
>  before I set to work.  I would:
>
> 1 Redefine viewHeight in  SpecCache to be windowSize/2 and cache the  
> spectral values,  rather than the displayed values.  This means I  
> will always  get a hit on the cache when zooming in and out  
> vertically, and  not redo all the ffts.
> 2 Remove all reference to max and min frequencies  in the spectral  
> calculations.
> 3 Move the  summing-for-display section of ComputeSpectrum (lines  
> 154-189)  into TrackArtist::DrawClipSpectrum about line 1435.  This  
>  would involve the max and min displayed frequencies.  It would get   
> used for horizontal scrolling however (it isn't currently) as the   
> displayed greyscale values would not be cached, so would  presumably  
> slow this down.
>
> Is this a good  idea?
>
> Martyn