Google has unveiled DolphinGemma, a new large-scale language model, which from the name can be guessed to be a dolphin language model! Dolphins are highly intelligent living beings with their own language to communicate among themselves, but it’s a challenge for humans to understand the language of dolphins. With the LLM technology coming in to help learn the patterns of dolphin language, DolphinGemma is a collaboration between Google and a team of researchers from Georgia Tech University and the Wild Dolphin Project (WDP), utilizing the sound and video footage archive of WDP accumulated since 1985 along with Google’s Gemma-based model. The trained DolphinGemma model has 400M parameters and uses Google’s SoundStream technique to convert dolphin sound waves into tokens.
Currently, WDP has started using the model in real-world applications with promising initial results, helping researchers to better understand dolphin communication. In addition to DolphinGemma, Google also has another project called CHAT (Cetacean Hearing Augmentation Telemetry) in collaboration with Georgia Tech, using Pixel smartphones as devices to listen to dolphin sounds underwater and generate simple dolphin sounds in terms of individual words like asking about objects in the water to communicate with dolphins (the first version of the device is Pixel 6 and is being upgraded to Pixel 9).
TLDR: Google introduces DolphinGemma, a collaboration with researchers to understand dolphin communication using a large-scale language model and innovative technology like SoundStream and Pixel smartphones for underwater communication.
Leave a Comment