there are many different cases and different technologies to achieve
automatic classification. And each technology works more or less well
depending on the type of searched sounds, their variability, the quality
and complexity of the recordings, the context, etc.
One technique is just image-matching, based on spectrographic images of
sample sounds to be searched within long recordings with more or less
flexible image matching algorithms. In some cases, it works. Another
technique is feature extraction of identified sounds, and then searching a
match with the features extracted form the recording to be searched. In
some cases it is possible to find very useful and "robust" features to be
searched.
There are also two different philosophies: one is based on known models to
be searched and on the training of an algorithm to search something that
matches those models, another searches for acoustic "objects" and then
tries to organize these objects into categories (usually clusters of
similar objects in a multidimensional space), finally, each group of
similar objects is compared with reference objects to possibly find
matches, and also to identify "unknown" objects. Here the main problem is
to define rules to identify and describe acoustic objects=85. and also how =
to
link objects to super-objects, e.g. notes organized into a defined song.
I think it would be very interesting to feed a debate about acoustic
classification=85.
Gianni
2013/7/17 wildambience <>
> **
>
>
> Thanks for the input Gianni, Bernie & Peter - it seems automating
> recognition of bioacoustics is obviously more complex than 'simply'
> matching an image (as is done in reverse-image searching)!
>
>
> --- In Gianni Pavan <>
> wrote:
> >
> > To date there are no systems able to automatically recognize species by
> > their sound. Many teams are working on this issue worldwide, for both
> > terrestrial and marine species. Systems available now work on limited
> sets
> > and require training with known species' sounds. In any case these
> require
> > supervision by human experts and careful tuning. Many systems work well
> in
> > the lab, but then, in the "real world" they fail because of the difficu=
lt
> > to predict variability of species' behavior, habitat propagation
> features,
> > reverberation, overlap of many sounds together, anthropogenic noises,
> etc.
> > etc.
> > However there are cases where algorithms work well in searching and
> > identifying specific sound events in huge recording sets.
> > This is the emerging field of computational bioacoustics and I'm sure i=
t
> is
> > worth the effort.
> > Our ears and eyes (to look at spectrograms) are still valid tools, in
> > particular if supported by our capability to use the context and a prio=
ri
> > knowledge to refine our findings.
> >
> > Gianni
> >
> >
> > 2013/7/17 Bernie Krause <>
>
> >
> > > The US Navy actually tried match-to-sample tech on their Trident
> > > subs=85vehicles that sometimes had over 3200 hydrophones embedded in =
the
> > > hull=85in the 1970s and 1980s. During that period they tried to ID wh=
ale
> vox
> > > and several hundred signatures of fish to compare to the signatures o=
f
> > > Soviet subs. Didn't work 'cause there were too many variations within
> each
> > > species and, altho, they still use something similar, it is still
> > > problematic in many instances. They could never overcome the problem
> > > introduced by Doppler shift=85objects moving at different speeds towa=
rd
> or
> > > away from each other.
> > >
> > > Bernie
> > >
> > >
> > > On Jul 16, 2013, at 6:29 PM, Peter Shute <> wrote:
> > >
> > > > I assume there are many people working on automatic species
> recognition,
> > > but I don't know how reliable it is yet. I thought they still require=
d
> > > verification by humans.
> > > >
> > > > But just the fact that recordings are being made and kept is a good
> > > thing. Maybe someone with as much resources as Google could do the
> initial
> > > analysis, and make the tagged sections of the recordings available fo=
r
> us
> > > to listen to. But imagine the outrage - first they photograph all of
> our
> > > houses, then they make continuous recordings.
> > > >
> > > > It's an interesting field, and it would be good if anyone here with
> > > knowledge of it could fill us in on it.
> > > >
> > > > Peter Shute
> > > >
> > > >> -----Original Message-----
> > > >> From:
> > > >> On Behalf Of wildambienc=
e
> > > >> Sent: Wednesday, 17 July 2013 9:11 AM
> > > >> To:
> > > >> Subject: [Nature Recordists] Article: "New Technology Tracks
> > > >> Species by Their Sounds"
> > > >>
> > > >>
> > > >>
> > > >> Interesting article from Nature World News -
> > > >> http://www.natureworldnews.com/articles/3001/20130716/new-tech
> > > >> nology-tracks-species-sounds.htm
> > > >>
> > > >> It got me thinking of the potential of this type of
> > > >> technology. As each species has unique calls which could be
> > > >> interpreted as unique visual patterns on a spectrogram, these
> > > >> visuals could be automatically scanned and interpreted to
> > > >> produce data on species distribution and abundance, much the
> > > >> same way as one can search by image on Google or Tineye.
> > > >>
> > > >> Snippet - "A new cyber infrastructure developed by scientists
> > > >> enables real-time acoustic recording and automatic species
> > > >> identification in remote locations of the world, offering
> > > >> anyone in the world quick and easy access of not only what
> > > >> creatures inhabit a given area, but how many of them there
> > > >> are - a key to measuring nature's response to on-going
> > > >> climate change and human invasion."
> > > >
> > > >
> > > > ------------------------------------
> > > >
> > > > "While a picture is worth a thousand words, a
> > > > sound is worth a thousand pictures." R. Murray Schafer via Bernie
> Krause.
> > > >
> > > > Yahoo! Groups Links
> > > >
> > > >
> > > >
> > > >
> > >
> > > Wild Sanctuary
> > > POB 536
> > > Glen Ellen, CA 95442
> > > 707-996-6677
> > > http://www.wildsanctuary.com
> > >
>
> > > SKYPE: biophony
> > > FaceBook:
> > > http://www.facebook.com/TheGreatAnimalOrchestra
> > > http://www.facebook.com/BernieKrauseAuthor
> > > Twitter:
> > > http://www.twitter.com/berniekrause
> > > YouTube:
> > > https://www.youtube.com/BernieKrauseTV
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > ------------------------------------
> > >
> > > "While a picture is worth a thousand words, a
> > > sound is worth a thousand pictures." R. Murray Schafer via Bernie
> Krause.
> > >
> > > Yahoo! Groups Links
> > >
> > >
> > >
> > >
> >
> >
> > --
> > Centro Interdisciplinare di Bioacustica e Ricerche Ambientali
> > Universit=E0 degli Studi di Pavia
> > Via Taramelli 24, 27100 Pavia
> > http://www.unipv.it/cibra
> > http://mammiferimarini.unipv.it
> >
> >
> >
>
>
>
--
Centro Interdisciplinare di Bioacustica e Ricerche Ambientali
Universit=E0 degli Studi di Pavia
Via Taramelli 24, 27100 Pavia
http://www.unipv.it/cibra
http://mammiferimarini.unipv.it
|