Methods - Soundscape Ecology: Principles, Patterns, Methods and Applications - page 244

This algorithm can also be applied to compare the differences between two
magnitudes along the axes of the frequencies comparing the magnitude of a
frequency bin with the magnitude of a successive frequency bin at the same time.
The combination of these two indices ACIt and ACIf becomes a better indicator
of the acoustic complexity inside a spectrogram.

9.7.2.6 ACI Evenness

This index measures the distribution of ACI along each frequency class

1
X pi 2

ACI e ¼

where pi is the relative abundance of the ACI( Δ fl ) index along the different

i frequency classes (Farina et al. 2012 ).
This index describes the level of saturation of a spectrogram to which all the
frequencies belong.

9.8 Pattern Recognition of Vocalizations

The identification of species has primarily a bioacoustics conservation valence, but
the identification of species from a sound file may also find several applications in
the field of soundscape ecology. In particular, the identification may be extremely
useful to validate synthetic (blind) indexes.
Vocal animals have species-specific signatures that can be extracted for their
identification. Today this is relatively easy for species that have stereotyped
vocalizations, adopting an ad hoc training set and after adequate filtering, but
when a complete soundscape is under investigation background noise produced
by wind or falling waters and the biophonic cacophonies produced by a low
acoustic partitioning create problems of identification and require great care
(Depraetere et al. 2012 ).
The automated identifications of animal vocalization using, for example, the
hidden Markov model (HMM), dynamic time warping (DTW), or Gaussian mixture
model (GMM), are procedures in quick improvement (Anderson et al. 1996 ; Kwan
et al. 2004 ; Chen and Maher 2006 ; Deecke and Janik 2006 ; Sumervuo et al. 2006;
Trifa et al. 2008 ), although at the present time these procedures can be used only for
a few species that have a simple sound structure and for records of quality without
the masking effects of geophonies (rain, wind) or anthrophonies (airplane noise,
traffic noise, etc.). For instance, Obrist et al. ( 2004 ) have applied a pattern recogni-
tion approach to the automated identification of most of the Swiss bat species using
“training-calls” (but see also Skowronski and Harris 2006 ).