\n
\n
![\"EarthSnap\"\/](\"https:\/\/www.europesays.com\/uk\/wp-content\/uploads\/2025\/05\/earthsnap-banner-news.webp.webp\")
\n<\/a><\/p>\n
Across decades of underwater recordings, blue whale songs appear as repeating acoustic signals embedded within vast, continuous streams of ocean noise.<\/p>\n
A team led by Ben Jancovich, a PhD candidate at the University of New South Wales (UNSW<\/a>) reshaped one call into many realistic versions. <\/p>\n The results showed that sparse evidence could still train a detector.<\/p>\n Jancovich\u2019s team built a software that flags target sounds, so old audio could reveal whale calls missed by earlier searches.<\/p>\n That approach points toward a bigger problem: rare animals often leave evidence long before people can interpret it.<\/p>\n One single call can train AI<\/p>\n Instead of collecting thousands of examples, the researchers copied one clean call and altered it in controlled ways.<\/p>\n Those edits used data augmentation, making changed copies to teach a system, by altering pitch, length, and background noise.<\/p>\n Altered copies taught the computer what natural variation and ocean distortion could do to the same basic song.<\/p>\n \u201cThe surprising outcome is that a relatively simple data augmentation process enables really good performance from that one single training example,\u201d said Jancovich.<\/p>\n Blue whales sing highly consistent songs<\/p>\n Blue whales were heavily hunted in the 20th century and their populations crashed. Some groups are slowly recovering, but they are still endangered.<\/p>\n NOAA Fisheries<\/a>, the U.S. agency for marine resources, reports that vessel strikes, entanglement, and ocean noise remain ongoing threats to blue whales.<\/p>\n Unlike many animals, groups of blue whales sing highly consistent songs, so one population often repeats one signature pattern.<\/p>\n That sameness gave the model a narrow target, because it only had to recognize a call with stable sound features.<\/p>\n For conservation work, reliable listening can help researchers track where whales travel without needing constant sightings.<\/p>\n Training without waste<\/p>\n Training large computer models can burn time, money, and electricity when every pattern starts from scratch.<\/p>\n Here, transfer learning, reusing a model trained for one task, allows the team to adapt software originally built for human speech.<\/p>\n A neural network<\/a>, a system built from layers that learn patterns, then compared whale-call shapes against the altered examples.<\/p>\n The finished system trained on a standard laptop within hours, which keeps the method within reach for smaller research groups.<\/p>\n Numbers that matter<\/p>\n For the Chagos pygmy blue whale, a smaller species tied to the central Indian Ocean, one-call training worked best.<\/p>\n A 99.4% recall score means the model found almost every call already present in the test set.<\/p>\n The share of flagged sounds that were true calls reached 91.2% after disputed labels were checked.<\/p>\n At 95.1%, the F1 score \u2013 one number balancing misses and false alarms \u2013 marked strong performance without proving perfection.<\/p>\n AI pinpoint meaningful sounds<\/p>\n Old recordings often come from passive acoustic monitoring, leaving recorders to listen unattended for months or years.<\/p>\n Underwater microphones called hydrophones<\/a> can capture ocean sound continuously, but the resulting files quickly become overwhelming.<\/p>\n Human analysts once inspected visual sound charts by hand, and fatigue could cause missed calls or inconsistent labels.<\/p>\n Automation does not remove expert judgment, but it can narrow the pile until humans review the most meaningful sounds.<\/p>\n Limitations of the method<\/p>\n Consistency makes the method powerful, yet that same requirement limits where it can work.<\/p>\n Dolphins<\/a>, for example, often use individual whistles, so one call would not represent enough of their vocal range.<\/p>\n A crowded whale chorus can also confuse the results, because many overlapping calls can resemble the target without clear boundaries.<\/p>\n Future versions may need examples of chorus and other difficult noise placed in the training set as rejected sounds.<\/p>\n Archives become evidence<\/p>\n Across the central Indian Ocean, the team plans to apply the detector to 25 years of recordings.<\/p>\n That long record could show whether whale songs changed over time, moved through seasons, or responded to human noise.<\/p>\n Because sound travels far underwater, acoustic archives can reveal animals that no ship, plane, or satellite happened to see.<\/p>\n Such records can turn scattered moments of listening into evidence about behavior, movement, and recovery.<\/p>\n Beyond the ocean<\/p>\n Forest microphones, desert recorders, and remote listening stations face the same problem as ocean archives.<\/p>\n Birds, insects, frogs, and other animals with repeatable calls could benefit if one strong recording can train a useful detector.<\/p>\n That possibility matters for rare species, especially when scientists hear them only a few times before conditions change.<\/p>\n \u201cIf accurate detectors can be trained from a single good recording, this can help us study rare and elusive species that have seldom been heard by humans,\u201d said Jancovich.<\/p>\n A single whale song will not solve every monitoring challenge, but it lowers the threshold for turning overlooked recordings into usable evidence.<\/p>\n Future tools will need rigorous validation, clearly defined limits, and open access to ensure researchers can apply them reliably at scale.<\/p>\n The study is published in the journal Nature<\/a>.<\/p>\n \u2014\u2013<\/p>\n Like what you read? Subscribe to our newsletter<\/a> for engaging articles, exclusive content, and the latest updates.<\/p>\n Check us out on EarthSnap<\/a>, a free app brought to you by Eric Ralls<\/a> and Earth.com.<\/p>\n \u2014\u2013<\/p>\n","protected":false},"excerpt":{"rendered":"Researchers have found that a single recorded blue whale call can be used to train an AI system…\n","protected":false},"author":2,"featured_media":936480,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3843],"tags":[728,70,16,15],"class_list":{"0":"post-936479","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-environment","8":"tag-environment","9":"tag-science","10":"tag-uk","11":"tag-united-kingdom"},"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@uk\/116514754345575722","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/posts\/936479","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/comments?post=936479"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/posts\/936479\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/media\/936480"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/media?parent=936479"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/categories?post=936479"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/uk\/wp-json\/wp\/v2\/tags?post=936479"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}