🎙️ Voice is AI-generated. Inconsistencies may occur.
Google has announced progress on an artificial intelligence model designed to analyze dolphin communication.
The AI, called DolphinGemma, was developed in collaboration with the Georgia Institute of Technology and the Wild Dolphin Project (WDP), and it seeks to decode the complex vocalizations of Atlantic spotted dolphins.
For decades, marine scientists have studied the intricate clicks, whistles and burst pulses used by dolphins, trying to understand whether these sounds form a kind of language.
DolphinGemma leverages advanced audio modeling and AI techniques to help researchers uncover patterns in the vocalizations, which could help scientists get closer to determining whether dolphin communications have structures similar to human language.
Google announced its progress on DolphinGemma on April 14—National Dolphin Day.
The AI model is not designed to translate dolphin speech directly but to recognize sequences and predict likely responses, functioning similarly to how human-focused language models anticipate the next word in a sentence.
The initiative marks a significant step in applying large language model technology to interspecies communication research.
The foundation for the research is a dataset collected by the WDP, a nonprofit in Florida that has conducted underwater studies of Atlantic spotted dolphins in the Bahamas since 1985. WDP's long-term, noninvasive approach has produced a rich collection of audio and video recordings tied to specific dolphins and behaviors.
These observations have shown key types of vocalizations: signature whistles used for reunification between mothers and calves, burst-pulse squawks associated with conflict, and buzzing clicks made during courtship or predator encounters. By correlating sounds with specific social contexts, WDP has laid the groundwork for DolphinGemma's pattern recognition.
The model uses Google's SoundStream tokenizer to efficiently process dolphin sounds and is small enough (about 400 million parameters) to run on Pixel smartphones, which researchers use in the field—allowing real-time analysis and interaction without the need for large-scale computing infrastructure.
WDP and Georgia Tech are also deploying the Cetacean Hearing Augmentation Telemetry (CHAT) system, a separate project that uses synthetic whistles to establish a simple, shared vocabulary between humans and dolphins. When dolphins mimic these artificial signals, CHAT identifies them and prompts researchers to respond with associated objects, reinforcing the association.
A Google spokesperson told Newsweek: "Though still early days, WDP's use of DolphinGemma in the field promises short and long-term advances in understanding Atlantic spotted dolphin communication, with potential to speed up research across, and maybe between, many different species as it becomes more widely available to researchers this summer."
WDP founder Dr. Denise Herzing and Google DeepMind Research Scientist Dr. Thad Starner wrote: "This approach in the quest for interspecies communication pushes the boundaries of AI and our potential connection with the marine world."
Google said it planned to release DolphinGemma as an open model in the summer, allowing researchers to adapt it for other cetacean species, such as bottlenose or spinner dolphins.
Update 4/16/25, 9.54 a.m. ET: This article has been updated with comment from Google.
