In a groundbreaking development announced on National Dolphin Day (April 14, 2025), Google has unveiled DolphinGemma, an innovative AI model designed to decode dolphin vocalizations. This collaborative project between Google, the Wild Dolphin Project (WDP), and Georgia Institute of Technology aims to unravel the mysteries of how dolphins communicate with each other—and potentially with humans.
A Four-Decade Foundation of Research
The Wild Dolphin Project, led by Dr. Denise Herzing, has been studying Atlantic spotted dolphins in the Bahamas since 1985, making it the world’s longest-running underwater dolphin research initiative. Their “In Their World, on Their Terms” approach has generated a rich dataset connecting sounds with specific dolphin identities, behaviors, and social contexts.
The WDP has documented various vocalizations including signature whistles that act like names, burst-pulse squawks often heard during conflicts, and click buzzes used during courtship or shark encounters. This extensive collection of labeled data provided the perfect foundation for applying artificial intelligence.
How DolphinGemma Works
DolphinGemma functions much like language models designed for human speech but focuses on dolphin vocalizations. Built upon Google‘s Gemma AI foundation, the system processes natural dolphin sounds—whistles, clicks, and burst pulses—and predicts subsequent sounds in their sequences.
The AI uses Google’s SoundStream technology to convert complex dolphin vocalizations into manageable “tokens” for analysis. This approach mirrors how language models predict the next word in human sentences, but applied to dolphin acoustic patterns.
The system’s compact design (approximately 400 million parameters) allows it to run directly on Google Pixel phones, making it practical for field research. This represents a significant advancement over traditional manual analysis methods, which require tremendous time and effort to identify patterns.
Project CHAT: Moving Toward Two-Way Communication
Alongside DolphinGemma, researchers have developed the Cetacean Hearing Augmentation Telemetry (CHAT) system. While DolphinGemma analyzes natural dolphin communication, CHAT takes a different approach by trying to establish a shared vocabulary with dolphins.
The system creates synthetic whistles associated with specific objects dolphins enjoy interacting with, such as sargassum, seagrass, or research toys. Researchers hope the naturally curious dolphins will learn to mimic these sounds to “request” specific items, potentially creating a simple back-and-forth exchange.
The technology can generate dolphin-like sound sequences, including whistles and burst pulses as shown in early testing of DolphinGemma.
Similar Posts
Pixel Power: Mobile Technology Meets Marine Research
The CHAT system relies on Google’s smartphone technology for real-time analysis. Previously using Pixel 6 phones, researchers plan to upgrade to Pixel 9 for field research slated for summer 2025, which will integrate speaker and microphone capabilities specifically designed for underwater use.
This technology streamlines what once required bulky custom hardware. The Pixel phones perform high-fidelity analysis of dolphin sounds in real-time while simultaneously running complex AI models and algorithms—crucial capabilities for responsive field research in open ocean environments.
The mobile phones help researchers:
- Hear dolphin mimics accurately amid ocean noise
- Identify specific whistles in real-time
- Alert researchers (via bone-conducting headphones) which object a dolphin might be “requesting”
- Enable quick human responses to potential communication attempts
Ethical Considerations
Despite technological excitement, the project raises important ethical questions. There are concerns about the risk of unintentionally training dolphins rather than truly understanding their natural communication. The fundamental question remains: Do dolphin vocalizations constitute a true “language” with the infinite complexity linguistic experts consider necessary?
Some zoologists remain skeptical about whether dolphin communication meets the technical definition of language. The research team acknowledges these concerns while continuing their careful, non-invasive approach.
Democratizing Research Through Open-Source
Google plans to release DolphinGemma as an open-source model in summer 2025, enabling researchers worldwide to adapt the technology for studying various cetacean species beyond Atlantic spotted dolphins. This could accelerate discoveries across marine mammal research, potentially benefiting conservation efforts for these intelligent animals.
The open approach reflects a growing trend of using AI tools to study animal communication more broadly, with similar efforts focusing on whales, primates, and birds. These technologies may provide new insights into animal social dynamics, emotional states, and responses to environmental changes.
As this technology continues to develop, it represents not just a tool for scientific discovery but a potential bridge between species—helping humans better understand and potentially communicate with some of Earth’s most intelligent non-human inhabitants.
