Agriculture Reference
In-Depth Information
A soundscape consists of a complex of specific sounds (e.g., birdsong, flowing
water, train whistle) of varying intensity depending on the source and the distance
from the sensor. These sounds can be used as signatures since they are repeat-
able. A soundscape can also consist of sounds that occur at different frequencies
(e.g., birdsong at higher frequencies, mechanical sounds at lower frequencies).
Quantifying either type of signal is difficult since multiple organisms may sing
simultaneously and their frequencies may overlap, making signature identification
difficult. On the other hand, some animals have simple sounds (e.g., spring peeper
frogs,
Pseudacris crucifer
) that can be readily quantified. Some organisms signal
at low frequencies (e.g., American crow,
Corvus brachyrhynchos
), and thus signal
at frequencies similar to those of some mechanical sounds, introducing exceptions
to the idea that biological sound can be separated from mechanical sound based
on frequency analysis. These problems have prompted research into pattern rec-
ognition to characterize entities in the soundscape (e.g., Reynolds and Rose 1995,
Anderson et al. 1996, Acevedo et al. 2009, Ranjard and Ross 2008, Brandes 2008,
Waddle et al. 2009, Kasten et al. 2010).
To date, most research on soundscapes has focused on understanding acous-
tic characteristics based on descriptive and qualitative analysis of sounds (Schafer
1977, Krause 1998). Quantitative methods to analyze the soundscape using fre-
quency extraction have been developed by Napoletano (2004) and Qi et al. (2008)
using spectrograms. Analysis begins with the creation of a spectrogram, which is
a time-varying spectral representation of an acoustic signal that can be visualized,
with frequency (in hertz or Hz) of an acoustic signal on the
y
-axis and time on the
x
-axis (Haykin 1991). To analyze a spectrogram image produced from a sound
recording, one can transform the recording to an image that can then be divided into
intervals (e.g., 1-kHz intervals) using image analysis software. The power level rep-
resented by the pixel values in each interval can then be summed to provide a value
for each frequency interval. This enables the signal power in each frequency inter-
val to be quantified. A more efficient method is to compute the total Power Spectral
Density (PSD in watts Hz
−1
) (Welch 1967) for each 1-kHz interval. This method
requires less computational time and eliminates the need to produce spectrogram
images and subsequently apply image analysis techniques to quantify the number
of pixels in each frequency interval. The specifics for these latter computations are
described in Kasten et al. (2012).
Soundscape Index
Napoletano (2004) found that mechanical sounds (anthrophony) mostly occur at
low frequencies (1-2 kHz), whereas most biological sounds (biophony) are preva-
lent above 2 kHz. Geophony (e.g., wind and rain) typically occurs across the entire
soundscape spectrum. We developed a Normalized Difference Soundscape Index
(NDSI) to separate biophony from anthrophony:
NDSI =
(
βα
βα
−
+
)
(
)
