What to know

  • Google has developed DolphinGemma, an AI model designed to decode and generate dolphin vocalizations.
  • The project uses decades of dolphin audio data from the Wild Dolphin Project and Georgia Tech.
  • DolphinGemma analyzes, predicts, and mimics dolphin sounds, aiming to establish two-way communication.
  • This initiative could transform our understanding of animal communication and interspecies interaction.

Google is training its artificial intelligence model, DolphinGemma, to interpret and generate dolphin vocalizations. This project is a collaboration with the Wild Dolphin Project, which has gathered extensive audio and video recordings of wild Atlantic spotted dolphins in the Bahamas, and the Georgia Institute of Technology. The goal is to translate underwater dolphin sounds into meaningful information and potentially enable a real dialogue between humans and dolphins.

DolphinGemma is built on Google’s SoundStream tokenizer and contains around 400 million parameters. It has been trained on one of the world’s largest databases of cetacean communication. The AI model can analyze the structure of dolphin sounds, detect repeating patterns, predict upcoming vocalizations, and even generate new sounds that closely mimic real dolphin speech. This approach is similar to how language models for humans predict words in a sentence, but here, the focus is on building a shared vocabulary between species.

The research follows two main paths. First, it decodes the natural communication of dolphins by analyzing whistles, squawks, and buzzes that are linked to specific behaviors. Second, it seeks to establish a shared vocabulary by creating synthetic dolphin sounds and associating them with specific objects or actions. This method allows researchers to test whether dolphins can learn new sound associations or teach humans the correct meanings.

DolphinGemma’s development is grounded in ethical research practices. The Wild Dolphin Project has prioritized non-invasive methods, ensuring that the dolphins are studied in their natural environment and on their own terms. The AI’s ability to identify, predict, and mimic dolphin sounds marks a significant step toward two-way interspecies communication. If successful, this technology could deepen our understanding of marine life and open new possibilities for communicating with other animals.

Via: androidcentral