Elephants talk with very low sounds called rumbles. Often, two or more elephants call at the same time, like a chorus.

Scientists studied many recordings of wild elephants in Kenya. A computer could guess the situation, such as a greeting, even from these group calls. When elephants met again after some time apart, their voices became more like each other. This may help the family feel close.

African elephants communicate with deep calls called rumbles, and about 60 percent of these calls overlap with other elephants' voices. Because overlapping sounds are hard to study, earlier research mostly looked at single calls. A team including Earth Species Project and ElephantVoices used machine learning to analyze recordings of wild adult females in Amboseli National Park, Kenya, collected from 1986 to 2022.

The computer could identify the situation, such as a greeting or a call to move on, from both single calls and choruses. In three situations, including greetings after being apart, the elephants' voices became more similar to each other. This may help strengthen family bonds. The study appeared in Scientific Reports.

Roughly 60 percent of African elephant rumbles are produced in choruses, with two or more animals calling at once. Most acoustic research has studied isolated calls instead, because overlapping voices are hard to separate. A study in Scientific Reports, led by Earth Species Project with ElephantVoices and the Amboseli Trust for Elephants, took on the choruses. It used recordings of identifiable adult females made in Kenya's Amboseli National Park between 1986 and 2022.

The researchers sorted rumbles into six contexts, from contact calls to greetings, and trained machine-learning models to classify them. The models identified the context better than chance from both single calls and choruses. A model trained only on single calls could classify choruses too, so the choruses seem to be built from individual calls that fit the situation. The study has clear limits.

All the recordings came from one population, some contexts had few examples, and accuracy dropped sharply for elephants and families the models had not heard before. The work links sounds to situations and does not translate an elephant language. One finding hints at something social. In three contexts, including greetings, the elephants' voices converged, which may help strengthen the bonds between them.