On April 14, 2025, National Dolphin Day, researchers from Google LLC, Georgia Tech, and the Wild Dolphin Project (WDP) unveiled DolphinGemma, an innovative AI model designed to analyze and generate dolphin vocalizations. This groundbreaking announcement marks a significant step towards understanding and potentially engaging in two-way communication with dolphins, a pursuit that has fascinated scientists for decades.
DolphinGemma is trained on extensive underwater video and audio data collected over decades from WDP’s long-term study of Atlantic spotted dolphins (Stenella frontalis) in the Bahamas. This model identifies patterns in dolphin vocal sequences and generates realistic dolphin-like sounds, aiming to decode the complexities of dolphin communication.
“By identifying recurring sound patterns, clusters, and reliable sequences, the model can help researchers uncover hidden structures and potential meanings,” Google representatives stated. The AI system is based on Google’s lightweight Gemma models and leverages SoundStream for audio representation, ensuring it is efficient enough to run on mobile devices.
With approximately 400 million parameters, DolphinGemma is compact enough to be deployed on waterproofed Pixel phones, which researchers use in the field. This allows for real-time analysis of dolphin sounds, a crucial factor in understanding the nuances of their communication.
Since 1985, the WDP has conducted the longest-running study of wild Atlantic spotted dolphins, gathering one of the largest datasets of dolphin communications globally. Their research has linked specific vocalizations to particular behaviors, such as signature whistles that help reunite mother and calf pairs, and burst-pulse squawks and click buzzes associated with fighting and courtship.
In addition to DolphinGemma, the WDP and Georgia Tech have developed the CHAT (Cetacean Hearing Augmentation Telemetry) system. CHAT utilizes synthetic whistles linked to objects like sargassum or scarves. When dolphins mimic these synthetic whistles, the system can recognize them and notify researchers via bone-conducting headphones, facilitating real-time interactions.
A Google Pixel 6 was used for the high-fidelity analysis of dolphin sounds during initial tests. The next iteration, based on the upcoming Pixel 9, will integrate advanced deep learning and template matching capabilities directly into the device, reducing the need for specialized hardware.
“We’re not just listening anymore. We’re beginning to understand the patterns within the sounds,” Google noted, emphasizing the importance of context in understanding any species. The goal of this research is to bridge the communication gap between humans and dolphins through data-driven insights and shared interaction systems.
The DolphinGemma model is set to be released as an open model in the summer of 2025. While it is currently trained on Atlantic spotted dolphins, researchers anticipate that it will also benefit studies on other cetacean species, although fine-tuning will be necessary for those with different vocal patterns.
“By providing tools like DolphinGemma, we hope to give researchers worldwide the means to mine their acoustic datasets, accelerate the search for patterns, and collectively deepen our understanding of these intelligent marine mammals,” Google stated. The implications of this research extend beyond mere curiosity; they could significantly enhance our understanding of dolphin communication and behavior.
The journey to deciphering dolphin communication is likely to be long and complex, but the introduction of CHAT and DolphinGemma brings researchers closer to achieving a two-way communication model. By grasping the patterns behind dolphin sounds and establishing a shared vocabulary, it may one day be possible for researchers to communicate with dolphins effectively.
Moreover, the increasing challenges faced by dolphins, such as marine noise pollution from fishing, trawling, and shipping, highlight the urgency of this research. Dolphins rely heavily on sound for communication, navigation, and foraging, making the understanding of their vocalizations even more crucial in a rapidly changing ocean environment.
As Google and the WDP continue their collaboration, the potential for breakthroughs in understanding dolphin communication grows. The hope is that by cracking the code of dolphin vocalizations, researchers can not only facilitate better communication between humans and dolphins but also foster greater awareness and protection of these remarkable creatures inhabiting our oceans.
