Google’s New AI Tries Talking Dolphin - and It’s Getting Pretty Good at It

For decades, scientists have recorded dolphin whistles and clicks, wondering if there was a deeper meaning. Now, AI might finally be cracking the code. On National Dolphin Day i.e. April 14, Google introduced its new AI Model, DolphinGemma to decode “Dolphin Communication.”

Also read: Free ChatGPT Plus for Students Until May 31 – Announces OpenAI

This initiative by the tech giant towards bridging interspecies communication is an uncommon move. DolphinGemma is developed to understand and generate dolphin-like vocalizations. Built in collaboration with Georgia Tech and the Wild Dolphin Project (WDP), the model analyses whistles, clicks, and burst pulses to detect communication patterns and generate realistic dolphin sounds.

WDP, the world’s longest-running underwater dolphin study since 1985, focuses on wild Atlantic spotted dolphins in the Bahamas. Their non-invasive approach “In their world, on their terms” has produced decades of audio-visual data, mapping specific vocalizations to behaviours like reunion calls, fighting, and courtship.

DolphinGemma, a 400M-parameter model inspired by Google’s Gemma and Gemini families, uses the SoundStream tokenizer to process sound sequences. It functions like a language model predicting the next sound—and can run on Google Pixel phones used in the field.

Also in use is CHAT (Cetacean Hearing Augmentation Telemetry), a two-way interaction system that links synthetic whistles to objects dolphins enjoy. Future Pixel devices, including the Pixel 9, will power real-time analysis and interaction.

DolphinGemma will be open-sourced in summer 2025, with potential for adaptation to other cetaceans bringing us closer to meaningful dialogue between humans and dolphins.