The Dolphins They are one of the most intelligent animals on the planet, and have been subject to scientific study for decades. But its complex communication system remains one of the main barriers to raise the understanding of these mammals. It is what it seeks to contribute Google With Dolphingemma, an artificial intelligence capable of decoding the language of dolphins with the intention of creating a communication system to talk to them.
The project is born from Google Deepmind, Google’s research laboratory, and is supported by Gemma, the open AI models that the company offers to researchers and scientists for use in different areas. It also relies on the years of research developed by the Wild Dolphin Project (WDP) organization. So, Dolphingemmareleased on April 14 on the occasion of the National Dolphin Day, is trained to Learn the structure of dolphins vocalizations and generate new sequences of sounds similar to these.
Google has used some of its audio technologies (the Soundstream Tokenizer) to represent the sounds of the cetaceans, and then has processed them under a model architecture with model with more than 400 million parameters. From there, Dolphingemma works as a human linguistic model, processing the sounds to identify and structure patterns, and then predict which are most likely in each sequence.
Wild Project Dolphin has been studying several generations of a specific community of a specific community of Motaded Dolphins of the Atlantic (Stenella frontalis) savages in the Bahamas. And so far they had discovered that the whistles are used by mothers and their young to gather, that the graznidos use them in fights, and that the click humming are present both in courtships and in shark persecutions.
But with Dolphingemma they can now try to understand the structures and meanings of these sounds with less effort, to Establish patterns and rules that shape a possible language. Over time, Google explains in a statement, researchers will be able to create synthetic sounds to refer to objects and establish a shared vocabulary for interactive communication.
In this regard, Google and Wild Dolphin Project have also been working together with the Technological Institute of Georgia to develop an underwater computer capable of interacting with animals. The system is called Chat (auditory increase telemetry for cetaceans) And, in its most recent version, it is able to create synthetic whistles from Dolphingemma data, to associate them with marine objects and elements.
Chat has been working thanks to Smartphone Google Pixel 6 To analyze the sounds in real time without resorting to a personalized hardware. Its reduced size and energy consumption allow to reduce the cost of the device and make it more manageable under water. With the arrival of the Google Pixel 9 in the future, you can simultaneously execute deep learning models and template coincidence algorithms.
Google’s plan goes to share Dolphingemma as a Open model this summerso that other researchers can study dolphins from more parts of the world. This expansion, the company points out, may require adjustments at the level of vocalization of the different species, and that is why they offer their technology under an open policy.
“By providing tools like this, we hope to give researchers around the world the tools to extract their own acoustic data sets, accelerate the search for patterns and collectively deepen our understanding of these smart marine mammals”They conclude.
