Particle.news

Download on the App Store

Google Deploys AI Model to Decode Dolphin Communication in Real-Time

DolphinGemma, developed with 40 years of Wild Dolphin Project data, aims to uncover the structure of dolphin vocalizations and enable two-way interaction.

Atlantic spotted dolphins
File photo of dolphins, taken in Sao Miguel, Portugal, in 2016.
Image
Image

Overview

  • Google has launched DolphinGemma, a 400-million parameter AI model designed to decode dolphin vocalizations, now operational on Pixel phones for real-time field analysis.
  • The AI model, created in partnership with Georgia Tech and the Wild Dolphin Project, builds on four decades of non-invasive research on Atlantic spotted dolphins in the Bahamas.
  • DolphinGemma identifies patterns in dolphin sounds, such as signature whistles and click buzzes, to explore the potential for interspecies communication.
  • The CHAT system, integrated with Pixel 9 phones for the upcoming research season, aims to establish a shared vocabulary with dolphins using synthetic whistles tied to specific objects.
  • Google plans to release DolphinGemma as an open model this summer, enabling broader applications for cetacean communication research worldwide.