Sara L. Heimlich

Brydes whale (balaenoptera edeni) sounds collected from autonomous hydrophones in the eastern tropical Pacific, 1999–2001

An array of seven autonomous hydrophones, deployed within 12 N 8S and 95 110W, continuously recorded the frequency band 1–110 Hz from November 1999 to November 2001. Four low frequency sound types were observed, similar to Brydes whale vocalizations previously reported [E. Oleson et al., Mar. Mammal. Sci. 19, 407–419 (2003)]. Two averaged 2.5‐s duration and were commonly high amplitude sounds that exceeded the sampling frequency. Both had a distinct tone averaging 18.6 and 25.4 Hz, respectively, several frequency‐modulated higher tones that resembled harmonics but were not, and often an initial short broad band pulse which overlapped the tonal components. Two were lower amplitude sounds that typically began with a single constant‐frequency upper tone, averaging 36.6 Hz, 1.1‐s duration, and 28.7 Hz, 1.1‐s duration, respectively. Sometimes, an overlapping lower tone followed 0.7 s later, lasting 1.6 s and averaging 21.4 and 15.3 Hz, respectively. Occasionally the upper and lower tones alternated two or more...

A pulsed-air model of blue whale B call vocalizations

Blue whale sound production has been thought to occur by Helmholtz resonance via air flowing from the lungs into the upper respiratory spaces. This implies that the frequency of blue whale vocalizations might be directly proportional to the size of their sound-producing organs. Here we present a sound production mechanism where the fundamental and overtone frequencies of blue whale B calls can be well modeled using a series of short-duration (<1 s) wavelets. We propose that the likely source of these wavelets are pneumatic pulses caused by opening and closing of respiratory valves during air recirculation between the lungs and laryngeal sac. This vocal production model is similar to those proposed for humpback whales, where valve open/closure and vocal fold oscillation is passively driven by airflow between the lungs and upper respiratory spaces, and implies call frequencies could be actively changed by the animal to center fundamental tones at different frequency bands during the call series.

Software for viewing and checking automatic detections.

Automatic detection of animal vocalizations is now used widely for handling long‐duration recording. Automatic detection methods inevitably make some errors—both false positive (wrong detection) and false negative (missed call) errors. Here a system is described for checking these errors. The MATLAB system “Osprey” allows viewing spectrograms, manipulating their parameters, and making various measurements of the displayed sounds. Another program, “checkDetections”, takes a log file that was output by an automatic detection software and systematically displays the detected sound in Osprey, allows a user to indicate whether the detection is correct, and then skips to the next detection in the log. This allows for rapid checking of detected sounds and calculation of the false‐positive (wrong‐detection) error rate. A second system, “checkMissedCalls”, displays random segments of sound in which no calls were found and allows the user to check whether there really were calls. This allows the user to estimate th...

Annotated odontocete recordings for developing automated detection and classification methods

A free archive has been developed for research in automatic detection/classification of cetacean sounds. The archive contains many datasets, with each dataset comprising recordings and metadata for a given species and geographic area. This archive differs from other sound archives in three important respects: (1) Recordings are annotated to indicate where in time and frequency the sounds of a given species occur. These annotations are done manually to remove the bias of any automatic detection system. (2) The archive deliberately includes poor‐quality recordings, recordings encountered in any realistic detection/classification application. (3) Since performance of detection/classification methods depends heavily on the SNR of target sounds, the archive includes each vocalizations SNR so performance can be effectively represented. Until recently, most recordings in the archive were of baleen whales [Mellinger and Clark, Applied Acoustics 67(11), 2006]. However, datasets for toothed whales and dolphins are...

A comparison of optimized methods for the detection of blue whale sounds

Paul Thompson and colleagues published one of the first long‐term studies of mysticete sounds [Thompson and Freidl, Cetology 45, 1–19 (1982)]. Thompson analyzed sounds manually, finding and tallying vocalizations to arrive at a view of seasonal occurrence. Today the detection and counting tasks are often done by computer, using various methods for pattern recognition. Here we examine and compare three such methods for detecting the sounds blue whales: matched filtering, which may work well because of the stereotypy of blue whale vocalizations; spectrogram correlation, which may work well for the same reason and also because of the noise removal that can be done with it; and a neural network, which has worked well in other contexts for detecting right whale calls. The methods are configured using optimization procedures specialized for each method, and the results are compared for vocalizations recorded at different signal‐to‐noise ratios. The optimized detectors are applied to SOSUS data to detect sounds ...

Extensible detection and classification in Ishmael

Ishmael is a bioacoustic analysis software package that has offers automatic detection of animal calls via a number of user-configurable detection methods. Ishmael now offers three significant improvements. The first is that users can download pre-configured detectors from within Ishmael for several cetacean species, including several mysticetes and several odontocetes. Each such detector is characterized with performance-measurement curves (receiver operating characteristic, detection error tradeoff, precision-recall) for a test dataset. Information on the signal-to-noise ratio of test datasets is also provided, since performance information is nearly meaningless without some indication of call clarity. The second is that Ishmael is extensible via a MATLAB interface that allows users to write their own detection plug-ins. Ishmael sends sound data and associated metadata to the plug-in, which analyzes the data and sends back detections and classifications for Ishmael to further handle as it would handle d...

Passive acoustic monitoring in the Northern Gulf of Mexico using ocean gliders

Although George Ioup did not use ocean gliders for passive acoustic monitoring, he recognized their value as platforms for PAM and encouraged others to use them. They function well as PAM platforms because (1) they move slowly, minimizing flow noise; (2) they have no propeller or continuously running machinery, minimizing motor noise; (3) they collect acoustic data nearly continuously; (4) they traverse the upper water column every few hours, measuring temperature and salinity as needed for calculating sound speed profiles and enabling accurate modeling of long-range acoustic propagation; (5) they can cover hundreds to thousands of kilometers in distance during a deployment, enabling them to monitor a large area and/or repeatedly monitor a smaller area; and (6) some models can dive to 1000 m, the depth at which some deep-diving cetaceans—sperm and beaked whales, frequent targets of PAM operations—forage and vocalize. Two models of gliders equipped with passive acoustic recording systems were deployed in t...

Application of density estimation methods to datasets collected from a glider

Ocean gliders can provide an inexpensive alternative for marine mammal population density studies. Gliders can monitor bigger spatial areas than fixed passive acoustic recorders. It is a low-noise, low-speed platform, easy to set up, maneuver, and transport on land, deploy, and recover. They can be deployed for long periods and report near real-time results through Iridium modem. Furthermore, gliders can sense the environmental conditions of the survey area, which are important for estimating detection distances. The main objective of this work is to evaluate the use of ocean gliders for population density estimation. Current methodologies developed for fixed sensors will be extended to these platforms by employing both simulations and real experimental data. An opportunistic preliminary sea trial conducted in June 2014 allowed for testing of a Slocum glider fitted with an inexpensive acoustic recording system comprising of two hydrophones connected to an off-the-shelf voice recorder installed inside the ...

Introduction to the special issue on methods for marine mammal passive acoustics

Automated methods for marine mammal passive acoustic research reach back at least 33 years, when Clark (1980) developed a direction-finding device to locate southern right whales. Automated methods soon became essential to the field, as rapid advances in technology and methodology led to huge volumes of passive-acoustic data, too large for convenient or timely manual analysis. By the early 2000s, automated passive acoustic research and monitoring had become widespread, used worldwide by academic, industrial, governmental, and independent researchers and agencies. Development of automated analytical tools for detection, classification and localization (DCL) of marine mammals from acoustic data was blossoming into an independent field of research. By 2003, the DCL research community had grown large enough to warrant the organization of the first International Workshop on Detection, Classification, and Localization of Marine Mammals using Passive Acoustics. The first workshop, organized by Francine Desharnais in Halifax, NS, Canada, focused on right whales (Eubalaena spp.). This workshop established the precedent of publishing a dataset before the meeting, a dataset to be used by workshop participants competing in the development of the best detection method. The result was the presentation of many novel techniques for detection and localization of right whale calls, and a broad education for all participants about the wide variety of research methods that might work best for this task. It also resulted in the publication of a special issue of Canadian Acoustics devoted to peer-reviewed papers from the workshop (Desharnais and Hay, 2004). The second workshop, organized by Olivier Adam and held in Monaco in 2005, focused on sperm whales (Physeter macrocephalus). Many of the practices of the first workshop were repeated, thus establishing several traditions for these workshops: Holding the workshop every other year (in oddnumbered years); alternating workshop locations between North America and Europe; publishing a workshop dataset some months before the workshop and holding a competition to see which methods had the best performance with this dataset; and publishing many of the results presented at the workshop in a special issue of a journal. The special issue related to the 2005 workshop appeared in Applied Acoustics (Adam et al., 2006). The third workshop, held in Boston in 2007 and organized by Dave Moretti, repeated and furthered the above traditions. Its focal species was Blainville’s beaked whale (Mesoplodon densirostris), and the corresponding special journal issue was in Canadian Acoustics (Moretti et al., 2008). The fourth workshop was organized by Gianni Pavan and held in Pavia, Italy in September 2009. It focused on Cuvier’s beaked whale (Ziphius cavirostris), with the special journal issue appearing in Applied Acoustics (Pavan et al., 2010). The fifth workshop was held at Mt. Hood, OR, in 2011, and was organized by the authors of this paper at Oregon State University. The focus was the detection and classification of odontocete sounds, including both whistles and clicks. This workshop was the first to focus on multiple components of odontocete vocalizations. It was at this workshop that the idea of a special issue of JASA devoted to methods for marine mammal passive acoustics was introduced and subsequently accepted. Since the first workshop 10 years ago, the field has changed in many ways. Machine learning methods, many derived from research in speech recognition or machine vision, have become widespread and prominent in detection and classification. Localization methods have incorporated acoustic propagation codes to obtain higher levels of localization accuracy. Methods for population density estimation (DE) have become important in the field, to the extent that the workshop is no longer “the DCL workshop” but rather has become “the DCLDE workshop.” Methods have changed from research topics into practical systems for everyday use. Implementations of detection and classification methods have been developed for networked systems capable of handling hundreds of hydrophones in real time. Many of these advances are related in the papers in this special issue, and we hope you find them interesting and informative. We, the guest editors, would like to thank the JASA editorial staff, particularly Allan Pierce, Editor-in-Chief for assistance in assembling and editing this special issue. We would like to thank the Office of Naval Research for supporting this special issue, and for longstanding support of the DCLDE workshops. We would also like to thank all of the authors who have contributed such high-quality articles that make this issue a truly special one.

Detecting humpback whale sounds in the Bering Sea: Confounding sounds in a cacophony of noise.

Humpback whales (Megaptera novaeangliae) are top predators of large zooplankton and forage fish, and one of the most common large whales in the Bering Sea. While present on feeding grounds, humpback whales produce nonsong sounds probably associated with feeding or social contacts. However, little is known about these highly variable sounds, and their detection is challenging. Recordings were collected during 2006–2007 at the long‐term oceanographic moorings M2, M4, and M5 in the eastern Bering Sea. Passive acoustic detection of humpback whale calls in these recordings was confounded by a variety of other sounds, which fall within the same parameters as nonsong humpback vocalizations. An automatic algorithm that detects tonal sounds in the 300–950 Hz frequency band was used to find humpback calls. Raw detections were visually examined to verify the accuracy of the detections. This algorithm resulted in a significant number of wrong detections (false positives), especially sounds of bearded seals (Erignathu...