Blue whales: An acoustic library helps us find what we can’t see

Reverberating through the ice shelves and gyres of the Southern Ocean are the undersongs of the largest animal that has ever lived on this planet, the Antarctic blue whale. Telling tales of the hunt for krill, of navigation and seduction, these tunes can carry for hundreds of miles.

And the world is listening: Moored around Antarctica is a loose ring of passive acoustic monitoring devices, or PAMs, deployed by various academic institutions. Released by oceanographic research vessels, the devices sink to the seafloor where they record a remote and often hostile realm that is practically out of reach of scientists. After about a year, a returning ship plays a coded message that trips a wireless trigger and frees the PAM recording to the surface.

“The idea here is that these animals are really rarely encountered on oceangoing voyages through visual surveys. There’s a small number of sightings of them, but they make really loud noises that can be detected over really large ranges,” says Brian Miller, a marine mammal acoustician at the Australian Antarctic Division in Kingston, Australia. 

After industrial whaling annihilated 99% of the blue whale population during the 20th century, the International Whaling Commission banned hunting in 1966. The critically endangered Antarctic blue whale is the largest of the species, with some reaching 110 feet in length and 330,000 pounds. Scientists estimate that about 3,000 individuals remain. And although blue whales may be well known, they are not well understood. 

PAMs are an excellent method to get to know them better, Dr. Miller says. “It seems like listening for these animals might be a better option for monitoring their recovery and their status ... than spending lots of money to send lots of ships with visual observers to look for them.”

Both sexes produce what researchers label “D” calls such as these, which are believed to be used for social contact, especially during foraging. The noises here are sped up eight times. Headphones are recommended. (Credit for this and subsequent audio: Brian Miller/Australian Antarctic Division)

When a PAM recording returns to its laboratory, analysts plumb its depths for sounds that can help determine the blue whale’s migratory paths and population, as well as new insights into its life. And yet, a year’s worth of sound also requires ... a year’s worth of listening.

To solve the prohibitive task of deciphering such a huge data set, cetologists have allied with computer scientists to develop custom software algorithms – instructions based on the patterns of previous whale sounds scripted into computer code – that can rapidly swim through oceans of data to home in on whale calls. Considering the constrained resources and diverse skills required for such work, these bespoke software packages may be only partially effective – or are poorly distributed among researchers.

So, to pool resources, unite international efforts, and present a broad range of recordings to test automated algorithms for the study of Antarctic blue whales (and fin whales, another species of concern), the scientific community has been broadening its efforts. The Acoustic Trends Working Group – funded through the International Whaling Commission’s Southern Ocean Research Partnership – has developed the online open-access Acoustic Trends Annotated Library. 

Dr. Miller, who served as project coordinator and lead author of the paper announcing the library, worked with colleagues to gather 2,000 hours of acoustic data from five nations, representing five separate years and instruments in each of the four circumpolar Antarctic regions. Next, a coterie of highly skilled analysts with extensive experience in identifying whale calls made 105,161 annotations identifying sound types, durations, and frequencies. Divided into hourlong blocks, the recordings were converted into standard .wav audio format and coupled with annotated text files. 

Fin whale “downsweep” notes: These repeated sounds form a song produced only by males during the breeding season.

While blue whale calls can exceed 180 decibels – louder than the roar of jet airplane – most of their sounds are infrasonic, meaning the frequency of their sound waves falls below the threshold of human hearing. So, to “hear” a blue whale, analysts watch for its sounds by feeding recordings into spectrogram software, which in turn renders sound signals into two-dimensional representations of waves across their computer screens.

“It is quite tricky,” says Emmanuelle Leroy, a bioacoustician who helped develop the digital library, about the annotating process. Variation in marking the boundaries and frequencies of calls is common between annotators and even for a single person when repeating work, she says. Improperly marking a sound can result in algorithms detecting false positives or missing whale sounds altogether. 

Because the library aims to provide the “ground truth” data set for scientists the world over to test their algorithms against, the team developed strict annotation parameters. Dr. Leroy says she enjoys looking at acoustic data, but the time she spent was taxing. “Of course, you dream of [annotating] during the night,” she laughs. “You see .wav vocalizations everywhere.”

Background interference, such as the weird whining and cracking of polar ice, is common in the recordings but also important in training algorithms to cut through the clamor of the sea and identify targeted sounds. “It comes down to diversity. If you want to analyze diverse data sets, you’ve got to train your algorithms on diverse data sets,” says Dr. Miller.

The acoustic library is already being used by researchers at the University of Concepción in Chile to compare blue and fin whale vocalizations over time and at different sites. And Shyam Madhusudhana, a postdoctoral fellow at Cornell University’s Lisa Yang Center for Conservation Bioacoustics, is using the library in the development of a global baleen whale sound detector, whereby researchers can simply drop in their own data sets to rapidly identify calls. He is applying deep-learning computer software to construct the detector. 

Already successful in automatic speech recognition, deep learning mimics the way humans learn. Employing layers of increasingly complex computing system nodes, called artificial neural networks, the technique synthesizes data in the form of continuous examples – as opposed to linear instruction – to arrive at an answer. This method, which should more accurately identify whale calls across distance and throughout turbulent soundscapes, requires huge caches of training data. Hence the whale acoustic library, which, Dr. Madhusudhana says, is “vital for this field.”

Unlike other cetaceans, such as the humpback whale with its assortment of higher-frequency moans, shrieks, and bellows – and which Dr. Miller calls “the jazz improv singers of the whale world” – blue whales produce sounds that are simpler and can be reduced to two classifications with a medley of underlying units and phrases. Only males sing, presumably to attract a mate, whereas both sexes produce downsweeping “D” calls, thought to be used for social contact. Dr. Miller refers to the blue whale sounds as the “rhythm section.”

The repetitive ”Z” call: There are only a few sounds in the Antarctic blue whale repertoire. This call (named “Z” for the look of its sound pattern rendered visually) rolls into a song sung only by males. 

And yet for all its weight and sonance, the animal remains an enigma. “Blue whales are really mysterious,” says Dr. Miller. “Hundreds of thousands of them were killed during whaling, and we learned so little about them. ... We still, to this day, don’t know where the breeding grounds of the Antarctic blue whales are.”

Dr. Leroy, who is now on Réunion Island in the Indian Ocean researching the vocal repertoires of local cetaceans, agrees about the mystery. While she says that acoustics offer vital tools for understanding the seasonal and geographic movements of the Antarctic blue whale, there is still much to learn. 

“We are blind,” says Dr. Leroy. “We see the songs, but we are blind.”