158
range which should cover the expected frequency range of the sounds of interest,
ideally with a flat sensitivity over a broad range of frequencies, but if that is not the
case, a proper calibration of the system will allow to correct the sensitivity as a function of frequency. The sampling frequency of the digital recorder which should be
at least double than the maximum frequency of interest in the signal; and the
dynamic range of the whole system (the range of signal’s amplitudes that the system
is able to process without being saturated or indistinguishable from the system noise
floor). For example, if we are interested in recording sounds from a NBHF species
we would need a digital recorder with a sampling frequency of at least 384 kHz
which would allow us to obtain recordings up to 192 kHz, and a hydrophone able to
record up to at least 160 kHz. However, if the main aim is to record the more intense
low frequency clicks from sperm whales we don’t need a digital recorded with such
high sampling frequency but we may want to look for a system with a large dynamic
range. This is because sperm whales’ clicks are one of the loudest sounds in the
oceans reaching up to 230 dB what can surpass the highest limit of the recording
system and the result would be saturated signals.
Once we have the digital signal from the sounds recorded, typically there is such
large amount of data obtained that the sound analysis can be very time consuming
and computationally-intensive. There is a big variety of pieces of software such as
Triton. Ishmael, Raven, Kaleidoskope, Avisoft SASLab, etc., and algorithms that can
be used to visualize the signals of interest. For example, long-term spectral averages
(LTSAs) can be calculated to avoid looking at each single audio file. LTSAs are longterm spectrogram with each time segment consisting of an average of hundreds of
spectra, and the resulting plots allow visualization of large time series data sets and
searching for and logging sounds of interest, like dolphin whistles or clicks for later
analysis. Once the acoustic signals are noticeable in the LTSA, such audio files can
be inspected more closely looking at each spectrum or time series.
After visually detecting and logging the acoustic events of interest we can use
some of the pieces of software mentioned before or others to extract the spectral,
temporal and energy properties of the signals. Alternatively, we may want to isolate
each single signal/call for its characterization using custom-routines in MATLAB
(Mathworks, Natick, MA) or R, to give some examples. Finally, once a large sample
size of the acoustic repertoire of a species has been handled, usually researchers are
interested in developing acoustic detectors, which are pieces of software that enable
to detect and extract the target sounds in an automatic or semi-automatic manner.
Hence, ideally acousticians will use one or more automatic or semi-automatic
detectors to extract the signals of interest and characterize their spectral and temporal properties.
Several species of odontocetes can be found in Argentinean waters, all of which
produce underwater sounds. Table 6.1 summarizes the available information on the
acoustic repertoire of odontocete species distributed in Argentinean waters and
around the western Antarctic Peninsula. There are several species for which there is
no local data or the existing data is scarce such as Bottlenose dolphins (Tursiops
truncatus), Common dolphins (Delphinus delphis), Killer whales, Dusky dolphins,
Peale’s dolphins, Burmeister’s porpoise (Phocoena spinipinnis), spectacled
M.L. Melcón et al.
range which should cover the expected frequency range of the sounds of interest,
ideally with a flat sensitivity over a broad range of frequencies, but if that is not the
case, a proper calibration of the system will allow to correct the sensitivity as a function of frequency. The sampling frequency of the digital recorder which should be
at least double than the maximum frequency of interest in the signal; and the
dynamic range of the whole system (the range of signal’s amplitudes that the system
is able to process without being saturated or indistinguishable from the system noise
floor). For example, if we are interested in recording sounds from a NBHF species
we would need a digital recorder with a sampling frequency of at least 384 kHz
which would allow us to obtain recordings up to 192 kHz, and a hydrophone able to
record up to at least 160 kHz. However, if the main aim is to record the more intense
low frequency clicks from sperm whales we don’t need a digital recorded with such
high sampling frequency but we may want to look for a system with a large dynamic
range. This is because sperm whales’ clicks are one of the loudest sounds in the
oceans reaching up to 230 dB what can surpass the highest limit of the recording
system and the result would be saturated signals.
Once we have the digital signal from the sounds recorded, typically there is such
large amount of data obtained that the sound analysis can be very time consuming
and computationally-intensive. There is a big variety of pieces of software such as
Triton. Ishmael, Raven, Kaleidoskope, Avisoft SASLab, etc., and algorithms that can
be used to visualize the signals of interest. For example, long-term spectral averages
(LTSAs) can be calculated to avoid looking at each single audio file. LTSAs are longterm spectrogram with each time segment consisting of an average of hundreds of
spectra, and the resulting plots allow visualization of large time series data sets and
searching for and logging sounds of interest, like dolphin whistles or clicks for later
analysis. Once the acoustic signals are noticeable in the LTSA, such audio files can
be inspected more closely looking at each spectrum or time series.
After visually detecting and logging the acoustic events of interest we can use
some of the pieces of software mentioned before or others to extract the spectral,
temporal and energy properties of the signals. Alternatively, we may want to isolate
each single signal/call for its characterization using custom-routines in MATLAB
(Mathworks, Natick, MA) or R, to give some examples. Finally, once a large sample
size of the acoustic repertoire of a species has been handled, usually researchers are
interested in developing acoustic detectors, which are pieces of software that enable
to detect and extract the target sounds in an automatic or semi-automatic manner.
Hence, ideally acousticians will use one or more automatic or semi-automatic
detectors to extract the signals of interest and characterize their spectral and temporal properties.
Several species of odontocetes can be found in Argentinean waters, all of which
produce underwater sounds. Table 6.1 summarizes the available information on the
acoustic repertoire of odontocete species distributed in Argentinean waters and
around the western Antarctic Peninsula. There are several species for which there is
no local data or the existing data is scarce such as Bottlenose dolphins (Tursiops
truncatus), Common dolphins (Delphinus delphis), Killer whales, Dusky dolphins,
Peale’s dolphins, Burmeister’s porpoise (Phocoena spinipinnis), spectacled
M.L. Melcón et al.