naturerecordists
[Top] [All Lists]

Re: Match-to-sample software?

Subject: Re: Match-to-sample software?
From:
Date: Sun, 8 Sep 2002 17:34:21 EDT
I don't think Bill Evans is on this discussion list. He could talk about th=
is
much more adequately than I can (the details of Bill's projects can be foun=
d
at <www.old bird.org>), but having worked with him the past two spring
seasons I do know something of the issues involved. Not only does no "voice=

recognition" software to identify bird vocalizations exist (except for a fe=
w
species used as "laboratory rats" to perfect the technique) but it almost
certainly never will. It isn't that the technology isn't available, it is,=

much of it developed by Bill. It is simply impractical. To accomplish such =
a
program, even just limited to the 900 species or so found in North America=

north of Mexico, would require hundreds of people working hundreds of
thousands of man hours. And the market for such a product? No where even
remotely near to covering the expense. Never mind the other 9000 species in=

the world, a significant portion of whose vocalizations are unknown and
unrecorded.

Bill has pioneered the study of nocturnal flight calls of migrating birds.=

Just this year he and Michael O'Brien published a CD ROM with the flight
calls of 250 species of North American migratory birds. Programs have been=

written to detect and record several species (and that's just for the fligh=
t
calls; most species of passerines may have 5-30 other vocalizations plus an=

infinite amount of individual variation). For the past three spring seasons=

Bill has monitored the nocturnal flight calls of Dickcissal, a species whos=
e
flight call is quite distinctive and relatively easy to write a program to=

recognize. We have a transect of 14 stations across south Texas that Bill's=

earlier work suggested probably intercepted the great majority of the world=
's
population of the species during its spring migration. It is also a species=

of concern and worth monitoring because of its habits on its wintering
grounds. Essentially the entire world population winters in northern South=

America, mainly in Venezuela, in what amounts to one enormous flock. Not
surprisingly it can be a serious agricultural pest in these areas as hundre=
ds
of thousands of birds may descend upon a small grain growing area. An
effective control campaign mounted by South American farmers could literall=
y
drive a relatively common species to the brink of extinction before we even=

knew it was in trouble.

The program we use starts up at a preset time (20:30), recognizes and recor=
ds
each Dickcissal flight call it "hears", and shuts down at 05:30. A typical=

night in late April may detect 1000 or more birds in a flight over one
location. Some of these will be false detections. Various sounds sufficient=
ly
close to the Dickcissal call in quality may be detected by the program, as=

may certain mechanical noises (especially rain drops hitting the mic
apparatus). After downloading a night's detections we examine the
spectrographs of each one. It is relatively easy to differentiate between a=

positive (Dickcissal) detection and a false one (Indigo Bunting calls,
certain notes in Purple Martin dawn song, some insect and/or frog noise,
mechanical noise, etc.). Bill also has a "sparrow and warbler" program.
Flight calls of 30 or more species of sparrows and warblers are so similar =
to
the human ear that they can't be identified in the field or from a recordin=
g.
The apparatus works similarly to that in the case of the Dickcissal except=

that potentially 30 odd species could theoretically be detected and recorde=
d.
These can be identified in the same manner. Each spectrograph can be compar=
ed
with known identity graphs and confidently (usually) be identified to speci=
es
and quantified.

This is an over simplified view of the process. The day when one can point =
a
microphone at a vocalizing bird and get a digital readout identifying it wi=
ll
never be a practical reality in my opinion.
John Arvin


________________________________________________________________________
________________________________________________________________________

<Prev in Thread] Current Thread [Next in Thread>
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU