Somewhere in the deep water off Dominica, a sperm whale surfaces after a foraging dive and fires off a rapid burst of clicks. To the human ear, it sounds like a stuttering zipper. To researchers from UC Berkeley and Project CETI, those clicks now look like something far more intricate: acoustic signals whose internal frequency structure mirrors the vowel sounds in human speech.
Two peer-reviewed studies, published in spring 2025 and building on years of fieldwork in the Caribbean, reveal that sperm whale vocalizations called codas carry layered information along at least two independent channels, timing and spectral content, simultaneously. The findings, which drew international attention after additional analysis was presented in April 2026, suggest these animals exercise deliberate control over the fine structure of their clicks, a trait scientists had associated only with human speech production.
Clicks that behave like vowels
Sperm whales generate clicks using a pair of structures called phonic lips, located inside the massive, oil-filled spermaceti organ that dominates their heads. Air forced through the lips produces a pulse that reverberates through the organ, and the resulting click can be astonishingly loud, upward of 230 decibels. Whales string these clicks into patterned sequences, or codas, during social interactions near the surface.
The first of the two studies, published in Nature Communications, analyzed 8,719 annotated codas recorded by the Dominica Sperm Whale Project and identified four timing-related features that vary systematically across vocalizations: rhythm, tempo, rubato (subtle timing fluctuations), and ornamentation (extra clicks inserted into a sequence). These features function like building blocks, combining in ways that expand the number of distinct coda types well beyond what earlier classification systems recognized.
The second paper pushed the analysis into new territory. Using linear predictive coding, a technique borrowed directly from human phonetics, researchers examined the frequency content within individual clicks and found at least two discrete, recurring spectral patterns. One category, labeled “a-coda,” displays a single low-frequency formant peak, according to the study published in Proceedings of the Royal Society B. Other codas show frequency transitions between peaks, resembling the way human diphthongs blend two vowel sounds into one syllable.
Lead researcher Gasper Begus, who directs the work from UC Berkeley’s Artificial Intelligence Research lab, has called these spectral categories “slow vowels” because the frequency shifts unfold over a longer timescale than in human speech. His interpretation, shared through a university release, is that the whales likely shape these patterns intentionally rather than producing them as incidental byproducts of their click-generation anatomy.
How the data was collected
Project CETI, a nonprofit initiative that combines marine biology with machine learning, gathered its recordings using a layered observation system: acoustic tags temporarily attached to individual whales, underwater hydrophone buoys, and aerial drones. That multi-sensor approach gave the team behavioral context alongside the audio, allowing researchers to link particular codas to group movements, dive profiles, and social interactions, even though the precise meanings of specific calls remain unknown.
The 8,719-coda corpus represents one of the largest annotated datasets of sperm whale vocalizations ever assembled. Its size, combined with systematic annotation and statistical modeling, gives the observed patterns real analytical weight. Bibliographic records for the research are indexed through PubMed, and the methodology is described in detail within both published papers.
Why scientists urge caution
Structure is not the same as meaning. The research team has been explicit that while the spectral and timing patterns show clear organization, no one yet knows what information the whales are encoding. Begus and his colleagues can demonstrate that codas vary in systematic ways, but connecting those variations to specific social functions, whether individual identification, group coordination during deep foraging dives, or emotional signaling, requires evidence the current dataset does not provide.
A researcher at Oregon State University’s Marine Mammal Institute, commenting on the findings in mid-April 2026, offered a pointed caveat: while the patterns are striking, drawing direct parallels to human language risks overstating what the data actually show. Whale clicks and human vowels are produced by entirely different anatomical systems, and the analogy, though useful for illustration, could mislead people into thinking scientists have decoded whale speech. They have not. It is more accurate to say sperm whales exhibit complex vocal structure than to claim they possess language in the human sense, which would require evidence of grammar, compositional meaning, and open-ended expression.
Cross-population variation is another open question. Sperm whale groups in different ocean basins use distinct coda repertoires, but the current vowel-pattern analysis draws primarily from the Dominica population. Whether the same spectral categories appear in Pacific or Indian Ocean clans, or whether each group develops its own “vowel inventory,” has not been tested with comparable rigor. Without parallel datasets from other regions, it is unclear whether the observed patterns reflect a species-wide capacity or a local cultural tradition.
What this means for ocean noise and conservation
The structural complexity revealed by these studies carries a practical implication that extends well beyond academic curiosity. If codas encode information through subtle frequency shifts, low-frequency noise from commercial shipping could mask exactly the signal components that matter most, particularly if key formant-like peaks overlap with the rumble of engine traffic. Researchers have not yet quantified that interference, but the finding makes the question more urgent and points toward the need for more detailed acoustic impact assessments in waters where sperm whale populations overlap with busy shipping lanes.
For context, sperm whales are listed as vulnerable by the International Union for Conservation of Nature, and populations in the Caribbean are among the most studied but also among the most exposed to vessel traffic and underwater noise. If their communication system turns out to be more fragile than simpler models predicted, conservation strategies may need to account for acoustic habitat quality, not just physical habitat protection.
Playback experiments and cross-population comparisons as next steps
The next phase of work will likely need to connect structural variation to behavioral outcomes: tracking whether whales that produce different spectral patterns behave differently in social or foraging contexts. That could involve pairing long-duration acoustic recordings with precise movement data from tags, as well as mapping individual whales’ social networks to see if particular coda types cluster within families or clans.
Experimental approaches, such as carefully controlled playback of codas with altered timing or spectral profiles, might help test whether whales perceive and respond to these differences. Comparisons with other cetacean communication systems, particularly the well-studied signature whistles of bottlenose dolphins and the complex songs of humpback whales, could also help clarify what is unique about sperm whale vocal structure and what reflects broader patterns in marine mammal cognition.
Until those links between structure and function are firmly established, the new research should be understood as a powerful map of the acoustic terrain of sperm whale communication. It tells us these animals are doing something remarkably sophisticated with their clicks. It does not yet tell us what they are saying.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
