Google AI Decodes Dolphin Communication, Runs on Pixel Phones

Google has launched DolphinGemma, an open-source artificial intelligence (AI) model designed to analyze and decode dolphin communication. 

This implementation is a great step towards understanding the complex vocalizations of marine mammals.

DolphinGemma, developed in partnership with Georgia Tech and the Wild Dolphin Project (WDP), analyzes clicks, whistles, and burst pulses, and can even generate dolphin-like sound sequences. 

Researchers hope this breakthrough will help determine if dolphin communication constitutes a form of language.

The AI model was trained on decades of data from the WDP, the world’s longest-running underwater dolphin research project.

Since 1985, WDP has been studying Atlantic Spotted Dolphins in the Bahamas using a non-invasive approach. This extensive dataset has provided DolphinGemma with a rich understanding of dolphin vocal patterns.

“By identifying recurring sound patterns, clusters, and reliable sequences, the model can help researchers uncover hidden structures and potential meanings within the dolphins’ natural communication, a task previously requiring immense human effort,” Google stated in its announcement.

 

Notably, DolphinGemma, containing approximately 400 million parameters, is compact enough to run on Pixel phones, enabling researchers to conduct real-time analysis in the field.

The model utilizes Google’s SoundStream tokenizer to process dolphin sounds and predicts subsequent sounds in a sequence, similar to how human language models predict the next word.

 

DolphinGemma is also integrated with the CHAT (Cetacean Hearing Augmentation Telemetry) system, which aims to establish a shared vocabulary with dolphins by associating synthetic whistles with objects they interact with.