[vox-tech] Marking Audio file based on Freq.

Bill Broadley bill at cse.ucdavis.edu
Tue Mar 11 10:29:25 PDT 2008


> I think this puts me little closer to what I'm after. All my boss wants 
> is to flag likely places where there might be a match and then use a 
> human to decide if it's the bird species we want. So I think one pass 

Hmm, I'd be kinda surprised if that works.  Birds are far from sine
wave generators.

On page 2 there's a graph of a guy doing a bird interpretation that
will give some idea of how complicated the resulting waveform is.

Birds, airplanes, delivery trucks, trains, dogs, wind, etc are all going to
be active at wide ranges of frequencies, you'd be lucky if birds were
the only ones active at any frequency.  If the imitation is any good it looks
like the peak intensity is around 1000 Hz and a lesser intensity across
an entire range up to 4000 Hz.

> with a fixed window size might work well, especially since I have an 
> example to work with. The push to make it a plugin for Audacity is so 
> that we have an easy to use interface for shuttling around the audio 
> file and listening to pieces, although I question it's ability to work 
> with 6 hour long files, I can also auto cut those with other apps before 
> loading.

On page2 the python program handles graphing the results for arbitrary
sized files and graphs it over time.  The tricky part part will be displaying
it at a resolution high enough to see a dozen short chirps over 3 seconds.

Armed with the graphs generated on page2 someone could likely browse
to the more interesting wave forms in the audio tool of your choice...
even an ipod.

I'd start with a 5 minute sample at a time and work up/downward as needed,
then every time the user wants to advance have the next N minutes displayed.


More information about the vox-tech mailing list