Limits...
Passive acoustic monitoring of the temporal variability of odontocete tonal sounds from a long-term marine observatory.

Lin TH, Yu HY, Chen CF, Chou LS - PLoS ONE (2015)

Bottom Line: The seasonal variation of whistle usage involved the previous three parameters, in addition to the diversity of whistle clusters.Our results indicated that the species and behavioral composition of the local odontocete community may vary among seasonal and diurnal cycles.The current monitoring platform facilitates the evaluation of whistle usage based on group behavior and provides feature vectors for species and behavioral classification in future studies.

View Article: PubMed Central - PubMed

Affiliation: Institute of Ecology and Evolutionary Biology, National Taiwan University, No. 1, Sec. 4, Roosevelt Road, Taipei 10617, Taiwan (R.O.C.).

ABSTRACT
The developments of marine observatories and automatic sound detection algorithms have facilitated the long-term monitoring of multiple species of odontocetes. Although classification remains difficult, information on tonal sound in odontocetes (i.e., toothed whales, including dolphins and porpoises) can provide insights into the species composition and group behavior of these species. However, the approach to measure whistle contour parameters for detecting the variability of odontocete vocal behavior may be biased when the signal-to-noise ratio is low. Thus, methods for analyzing the whistle usage of an entire group are necessary. In this study, a local-max detector was used to detect burst pulses and representative frequencies of whistles within 4.5-48 kHz. Whistle contours were extracted and classified using an unsupervised method. Whistle characteristics and usage pattern were quantified based on the distribution of representative frequencies and the composition of whistle repertoires. Based on the one year recordings collected from the Marine Cable Hosted Observatory off northeastern Taiwan, odontocete burst pulses and whistles were primarily detected during the nighttime, especially after sunset. Whistle usage during the nighttime was more complex, and whistles with higher frequency were mainly detected during summer and fall. According to the multivariate analysis, the diurnal variation of whistle usage was primarily related to the change of mode frequency, diversity of representative frequency, and sequence complexity. The seasonal variation of whistle usage involved the previous three parameters, in addition to the diversity of whistle clusters. Our results indicated that the species and behavioral composition of the local odontocete community may vary among seasonal and diurnal cycles. The current monitoring platform facilitates the evaluation of whistle usage based on group behavior and provides feature vectors for species and behavioral classification in future studies.

Show MeSH
Example of automatic tonal sound detection and unsupervised classification.(a) Spectrograms produced from MACHO recording using fast Fourier transform with a Hamming window. (b) Burst pulses (red dots), harmonics (blue dots), and representative frequencies (black dots) obtained by the local-max detector. (c) Whistle contours were extracted using the pitch-tracking algorithm; different contours were labeled with different numbers. (d–g) The four whistle types were classified using the unsupervised method.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4414466&req=5

pone.0123943.g002: Example of automatic tonal sound detection and unsupervised classification.(a) Spectrograms produced from MACHO recording using fast Fourier transform with a Hamming window. (b) Burst pulses (red dots), harmonics (blue dots), and representative frequencies (black dots) obtained by the local-max detector. (c) Whistle contours were extracted using the pitch-tracking algorithm; different contours were labeled with different numbers. (d–g) The four whistle types were classified using the unsupervised method.

Mentions: The acoustic recordings collected from October 2011 to September 2012 were examined using the automatic detection and classification algorithm, a Matlab (MathWorks, Natrick, MA)-based program, developed by Lin et al. [15,29]. This algorithm includes 4 steps: automatic detection of tonal sounds, separation of burst pulses and harmonics, contour extraction, and unsupervised classification (Fig 2).


Passive acoustic monitoring of the temporal variability of odontocete tonal sounds from a long-term marine observatory.

Lin TH, Yu HY, Chen CF, Chou LS - PLoS ONE (2015)

Example of automatic tonal sound detection and unsupervised classification.(a) Spectrograms produced from MACHO recording using fast Fourier transform with a Hamming window. (b) Burst pulses (red dots), harmonics (blue dots), and representative frequencies (black dots) obtained by the local-max detector. (c) Whistle contours were extracted using the pitch-tracking algorithm; different contours were labeled with different numbers. (d–g) The four whistle types were classified using the unsupervised method.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4414466&req=5

pone.0123943.g002: Example of automatic tonal sound detection and unsupervised classification.(a) Spectrograms produced from MACHO recording using fast Fourier transform with a Hamming window. (b) Burst pulses (red dots), harmonics (blue dots), and representative frequencies (black dots) obtained by the local-max detector. (c) Whistle contours were extracted using the pitch-tracking algorithm; different contours were labeled with different numbers. (d–g) The four whistle types were classified using the unsupervised method.
Mentions: The acoustic recordings collected from October 2011 to September 2012 were examined using the automatic detection and classification algorithm, a Matlab (MathWorks, Natrick, MA)-based program, developed by Lin et al. [15,29]. This algorithm includes 4 steps: automatic detection of tonal sounds, separation of burst pulses and harmonics, contour extraction, and unsupervised classification (Fig 2).

Bottom Line: The seasonal variation of whistle usage involved the previous three parameters, in addition to the diversity of whistle clusters.Our results indicated that the species and behavioral composition of the local odontocete community may vary among seasonal and diurnal cycles.The current monitoring platform facilitates the evaluation of whistle usage based on group behavior and provides feature vectors for species and behavioral classification in future studies.

View Article: PubMed Central - PubMed

Affiliation: Institute of Ecology and Evolutionary Biology, National Taiwan University, No. 1, Sec. 4, Roosevelt Road, Taipei 10617, Taiwan (R.O.C.).

ABSTRACT
The developments of marine observatories and automatic sound detection algorithms have facilitated the long-term monitoring of multiple species of odontocetes. Although classification remains difficult, information on tonal sound in odontocetes (i.e., toothed whales, including dolphins and porpoises) can provide insights into the species composition and group behavior of these species. However, the approach to measure whistle contour parameters for detecting the variability of odontocete vocal behavior may be biased when the signal-to-noise ratio is low. Thus, methods for analyzing the whistle usage of an entire group are necessary. In this study, a local-max detector was used to detect burst pulses and representative frequencies of whistles within 4.5-48 kHz. Whistle contours were extracted and classified using an unsupervised method. Whistle characteristics and usage pattern were quantified based on the distribution of representative frequencies and the composition of whistle repertoires. Based on the one year recordings collected from the Marine Cable Hosted Observatory off northeastern Taiwan, odontocete burst pulses and whistles were primarily detected during the nighttime, especially after sunset. Whistle usage during the nighttime was more complex, and whistles with higher frequency were mainly detected during summer and fall. According to the multivariate analysis, the diurnal variation of whistle usage was primarily related to the change of mode frequency, diversity of representative frequency, and sequence complexity. The seasonal variation of whistle usage involved the previous three parameters, in addition to the diversity of whistle clusters. Our results indicated that the species and behavioral composition of the local odontocete community may vary among seasonal and diurnal cycles. The current monitoring platform facilitates the evaluation of whistle usage based on group behavior and provides feature vectors for species and behavioral classification in future studies.

Show MeSH