- Google and the Wild Dolphin project have developed a trained model to understand the vocalizations of the dolphins
- Dolphingemma can work directly on pixel smartphones
- Will be open source this summer
During most of human history, our relationship with Dolphins has been a unilateral conversation: we spoke, squeak and settle how we understood each other before throwing a fish. But now, Google has a plan to use AI to unite that division. Working with Georgia Tech and The Wild Dolphin Project (WDP), Google has created Dolphingemma, a new model of the IA trained to understand and even generate dolphins talk.
The WDP has been collecting data on a specific group of spotted dolphins from the wild Atlantic since 1985. The Bahamas headquarters has provided large amounts of audio, video and behavior notes as researchers have observed them, documenting each shower and buzz and trying to rebuild what everything means. This audio treasure is now feeding with Dolphingemma, which is based on the family of Google Open Gemma models. Dolphingemma takes dolphin sounds as an entrance, processes them using audio tokenizers such as Soundstream, and predicts what vocalization could come next. Imagine autocomplete autocomplete, but for dolphins.
The model is very thin and can be executed in a Google pixel. WDP is already implementing dolphingemma in the field this summer, using Pixel 9s on waterproof platforms. These models will listen, identify vocal patterns and help researchers mark significant sequences in real time.
Attend
Flipper speaks
But the final objective here is not just a passive listening. WDP and Georgia Tech are also working on a system called Chat (abbreviation for telemetry to increase cetacean audition), which is essentially a bidirectional communication system for humans and dolphins. The chat allows researchers to assign synthetic whistles to objects such as dolphins, including marine and floating steps scarves, and then wait to see if the dolphins mimic those sounds to request them. It is like inventing a shared language, except with underwater microphones instead of flash cards.
Dolphingemma not only analyzes the sounds of the dolphins after the fact; It helps anticipate what sounds could come, which allows faster response times and softer interactions. In essence, it is like a predictive keyboard for dolphins. The entire project is still at an early stage, but Google plans to open -source for the open source at the end of this year to accelerate progress.
The initial model is trained in the vocalizations of the stained dolphins of the Atlantic, but theoretically could adapt to other species with some tuning. The idea is to deliver to other researchers the keys to AI so that they can apply it to their own acoustic data sets. Of course, this is still a long way from chatting with dolphins about philosophy or their favorite snacks. There is no guarantee that the vocalizations of dolphins are assigned perfectly to human language. But Dolphingemma will help examine years of audio for significant patterns.
Dolphins are not the only animals that humans can use AI to communicate. Another group of scientists developed an AI algorithm to decode the emotions of pigs based on their grunts, shrieks and blackouts to help farmers understand their emotional and physical health. However, dolphins are undeniably more charismatic. Who knows, maybe one day you can ask for instructions from a dolphin while navigating, at least if you don’t let your phone fall into the water.
