>Generic software would have to be produced that could be "loaded" with the
>vocalizations from a given area, to take into account the suite of species
>present and the local dialects of many of them, but that doesn't seem too
>difficult.

I'm not so sure that voice recognition software for birds would be based on
individual vocalizations, or that every possible call would need to be
programmed in . One of my friends worked in a birdsong lab at Caltech, and
it seemed to me that like human voice recognition software, birdsong
recognition is based more on the quality of the voice (including things that
humans can't hear) than on a predefined pattern of notes that must occur for
the bird to be recognized. For example, I believe human voices can be
matched without all possible words that person might say being programmed in
(now, if you want the computer to UNDERSTAND what you're saying, that's an
entirely different matter!). Similarly, their technique was to look at
qualities of the voice pattern itself, along with certain "diagnostic" parts
of a call. So, while some songs may vary, it seems that there are pieces or
qualities that remain the same. This is an obvious application for neural
net computers, which are currently being programmed to recognize such things
as faces (even though one person's face is never exactly the same any two
times you see it). A computer based on a neural net program would recognize
the birdsong in approximately the same way you or I do in the field... For
example, Dennis may be able to recognize a birdsong as a fox sparrow in some
other part of the world, even though he has never heard that particular
dialect of fox sparrow. Neural nets are able to accomplish similar
functions (this same friend of mine now works at Oxford on neural net
applications).

Teresa Michelsen
temmi461 at ecy.wa.gov.