Hi recordists:
SOFTWARE TO PRODUCE SEPARATE SPECIES RECORDINGS from a complex natural
soundscape:
Has anybody any progress to report on using the public domain software
methods we have discussed before (Navy lab in MD, called source isolation
by simulated annealing) which uses a multi-track linear mic array? This
analysis is said to produce audio output as separate channels for each and
every different sound source in the original recording. I am still looking
for an affordable multi-track field recording system (6 - 8 tracks?) on
which to get some field data on which to try it out. The actual specs for
the RIFF type WAV file will allow any number of tracks (by frames) up to
256 (?) within a WAV file. In practice software has ONLY been written, to
my knowledge, using 1 or 2! Why have folks limited such a terrific file
type to such a narrow range of choices?
The simulated annealing software has yet to be written, but the algorithms
I can get my hands on. The effect should be spectacular if and when we get
it all to work. Some speculations below:
You record with six omni mics (from four to many), spaced, say, 1 foot
apart on a special mount on a tripod. This should be high rate
SYNCHRONIZED sampling digital, if possible. Many multi-channel digitizers
sequence through the channels, sampling at a series of closely-spaced
times, but although cheaper to build, this will not work for this analysis.
What you might get back is twenty separate channels, each BETTER than could
have been done at the time with perfectly aimed parabola mics, pointed at
each sound source (including the hidden birds that only called once during
the cut). Say, one from - 32 degrees, 30 meters of a Rufous Hummingbird,
one from -48 degrees and 20 meters of a Rufous Motmot calling once, one
from +40 degrees and 150 meters of a Keel-billed Toucan grating its low,
frog-like call, one from 5 degrees 18 meters of the Pied Puffbird that
rattled it feathers once, one from -80 degrees 5 meters of your companion,
who coughed in the middle of it, one from +38 degrees 40 meters of the
Orange-chinned Parakeet shrieking at its small flock, etc. The degrees
reference straight ahead, 90 deg to the array line, as zero degrees. Plus
means to the right of center, minus to the left. Up down axis is ignored
(so far).
There are so many practical questions this scenario raises - what about
non-directional sounds (essentially including also all sounds below the
array cut-off frequency - I would GUESS to be where the wavelength
approaches 2 x the array length, here 6 ft, of lambda = 12 ft, or
1000/12 ~~ 80 Hz. My guess is also these should be filtered out before
initial digitizing, but who nows.
How good is the distance part of the analysis? Can it resolve and produce
separate outputs for two birds, one at 20 meters and one at 30?
How does it handle common channel identical sounds - as: low airplane rumbles?
How will it deal with 300 insects of 3 different species, chorusing in the
surrounding canopy?
my very best,
Marty Michener
MIST Software Associates
75 Hannah Drive, Hollis, NH 03049
coming soon : EnjoyBirds bird identification software.
________________________________________________________________________
________________________________________________________________________
|