Skip to content
Artificial intelligence. Clearly. RSS feed
AI NEWS AI

News. Evidence. Perspective.

ResearchAnalysis

Elephant choruses grow more alike as social exchanges unfold

Machine learning found shared acoustic patterns in 902 elephant recordings from Kenya, offering a way to study collective communication without separating every overlapping voice.

An elephant chorus is more than several animals talking at once. In recordings from Kenya’s Amboseli National Park, machine-learning models could identify the social setting from overlapping rumbles—and some exchanges grew more acoustically alike as they continued. The finding gives researchers a new way to examine communication at the level of the group.[1][2]

The team analysed 902 recordings associated with 84 wild African savannah elephants from 39 families, collected during several periods between 1986 and 2022. The archive linked sounds to observed encounters, allowing the researchers to compare what a chorus sounded like with what the elephants were doing.[1][2]

Around 60% of elephant rumbles occur in choruses, according to the researchers’ background account. Some frequencies lie below ordinary human hearing, and when two animals vocalise together their signals interfere. A machine-learning system can search for statistical characteristics of the combined sound even when assigning every frequency to an individual elephant is difficult.[1][2]

Six occasions to rumble

The team investigated six behavioural settings: contact calls between separated elephants; attempts to initiate movement, described as let’s-go rumbles; cadenced exchanges occurring while feeding, resting or travelling slowly; coos directed by females toward calves; small greetings between close associates; and more elaborate greetings during reunions. These labels arose from observations of the animals, not from a computer discovering the meanings of previously unknown elephant words.[1][2]

Acoustic features allowed models to predict the behavioural context of a single rumble and, importantly, a chorus. Models trained using lone calls could also predict the setting of chorused calls. That result suggests individual and collective exchanges share some context-dependent features, rather than becoming utterly different kinds of sound when two animals overlap.[1][2]

Then came a more surprising result. In cadenced, contact and greeting exchanges, the rumbles tended to grow more acoustically similar as the interaction progressed. That is not the same as everyone making an identical noise. It means measurable properties of distinct sounds moved closer together over the course of certain social encounters.[1][2]

When individual calls become a group signal

The authors suggest this convergence could help coordinate a shared action, such as a group starting to move. That interpretation remains a hypothesis: no playback experiment tested whether changing the convergence of calls altered the elephants’ behaviour.[1][2]

The technical problem extends beyond elephants. Traditional acoustic classification often assumes that a target voice can be isolated from background noise. Social animals do not always respect that assumption. Treating the chorus as the object of analysis may reveal properties that vanish if the overlapping portion is simply discarded as unusable data.[1][2]

There is also an archival lesson. Decades of natural-history field recording can acquire new value when analytical methods improve, provided the original observations still connect sound to the circumstances in which it was produced. A recording alone is one kind of evidence; a recording linked to a specific social encounter is a far richer one.[1][2]

An archive becomes a new experiment

A useful next test would connect the timing of convergence with the timing of a group’s movement. Synchronised video and identification of the contributing animals could reveal whether one elephant leads the acoustic change or several animals adjust together. Playback experiments could then test whether the pattern influences the listeners.[1][2]

The analytical advance is to keep the overlap. A chorus that defeats an attempt to isolate each voice can still carry measurable information about the whole encounter. For researchers listening to social animals, that changes which recordings count as usable evidence.[1][2]

Follow the evidence

Sources & further reading

  1. Social context information and acoustic convergence in elephant vocal choruses Scientific Reports · 1 Oct 2026 · Accessed 24 Oct 2026
  2. Zoology: Elephant rumbles get herds on the same wavelength Nature Portfolio · 2 Oct 2026 · Accessed 26 Oct 2026