Is
this the way to go with COG surveys? What do COG's audio experts think?
Tony
ScienceDaily (July 28, 2008) — Switch on the mike,
start the recording, the stage is set for the local fauna!
Computer
scientists from the University of Bonn, in conjunction with the birdsong
archives of Berlin’s Humboldt University, have developed a kind of
‘Big Brother’ for birds. This has nothing to do with entertainment,
but a lot to do with the protection of nature. The new type of voice detector
involved can reliably recognise the characteristic birdsong of different
species of birds, thereby facilitating surveys of the bird population.
Europe’s
forests are falling silent as countless species of birds go on the red list of
endangered species. Yet in fact no-one can say what the exact position is with
some species. So as to have a reliable count of the territories of indigenous
birds it would practically be necessary to send out a whole horde of spare-time
ornithologists to count the birds. What is more, since the birds are often
hidden in the undergrowth or the tree tops, ornithologists need to rely on
their ears and their specialist knowledge. This means that in many areas it is
wellnigh impossible to map the bird population comprehensively and
continuously.
In
view of such problems environmental protection has to fall back on new
technical methods. Some of these are now being provided by Bonn scientists.
Computer scientists from the University of Bonn have developed detectors which
can recognise birdsong automatically. What this implies is that in the
preliminary stage microphones are placed at selected points in the wild; these
record all the sounds made, in some cases over a period of months. The new
computer software can then sift through the many hundreds of hours of recorded
material overnight and say how many birds of which species have been singing
and how often they have been doing this.
In his
project Daniel Wolff of the Institute of Computer Science at the University of
Bonn initially concentrated on the bio-acoustic recognition of the Savi’s
warbler and the chaffinch. He listened carefully to the various types of
birdsong, scrutinised them in a spectrogram and transferred the characteristics
to algorithms. As soon as specific parameters are met, the programme kicks in.
‘For example, the signal of the Savi’s warbler has a mean frequency
of 4 kHz, which is very typical. If, in addition, individual elements of the
signal are repeated at a frequency of 50 Hz, this is detected as the call of a
Savi’s warbler,’ Daniel explains. The chaffinch detector also
analyses periodic repetitions of elements like these. In doing so it reveals
more of a typical verse structure than the pitch of the chaffinch’s song.
The Savi’s
warbler detector, particularly, which was subjected to long-term monitoring at
Brandenburg’s Parsteiner Weiher, is characterised by what researchers
call ‘robust recognition’, i.e. a high degree of reliability.
Despite interference from rain, wind and amphibians the programme recognised,
with a 92% detection accuracy, the song of a species of bird which is still
found on the shores of the Baltic but which has become rare elsewhere in
Europe.
The
birdsong detectors are as yet only calibrated for the birdsong of individual
species. However, in the near future, Daniel Wulff thinks, it will be possible
to link them up to a kind of superdetector which can recognise as many species
as possible and, in combination with GPS coordinates, will make the mapping of
bird populations simpler and more efficient.
The
research field of bio-acoustics, he adds, is currently experiencing a boom.
Although it was in the 1970s that the first attempts were made ‘to detect
the chaffinch with much slower computers,’ Daniel says, with a nostalgic
smile, ‘what is decisive is that it’s only now that we are in a
position to store a large amount of recorded sound and place compact technology
in nature which can really run for months, e.g. with solar energy.’
Adapted from materials provided by Bonn, Universitaet, via AlphaGalileo.