DolphinGemma logo

DolphinGemma

Paid

Google AI model for decoding dolphin communication

4.5
Type
Saas
Company
Google

About DolphinGemma

DolphinGemma is a large language model developed by Google to aid in decoding dolphin communication. Built on insights from Gemma, Google's lightweight open model family, DolphinGemma uses the SoundStream tokenizer to process dolphin vocalizations and is designed as an audio-in, audio-out model. With approximately 400 million parameters, it is optimized to run on Pixel phones for field research. Trained on decades of acoustic data from the Wild Dolphin Project's study of Atlantic spotted dolphins (Stenella frontalis), the model learns the structure of dolphin sounds and can generate novel dolphin-like sound sequences. This collaboration with Georgia Tech and the Wild Dolphin Project aims to advance interspecies communication by analyzing natural dolphin vocalizations and producing realistic responses.

Key Features

Audio-in, audio-out model for dolphin vocalizations
~400 million parameters, optimized for on-device use on Pixel phones
Uses SoundStream tokenizer to represent dolphin sounds
Built upon Gemma, Google's lightweight open model family
Trained on decades of labeled acoustic data from wild Atlantic spotted dolphins
Collaboration with Wild Dolphin Project and Georgia Tech

Pros & Cons

Pros
  • Leverages decades of high-quality field data from Wild Dolphin Project
  • Lightweight model runs on consumer smartphones for real-time field use
  • Built on open Gemma framework, enabling further research
  • Non-invasive approach respects dolphins' natural behavior
Cons
  • Still in early research phase, not yet proven to understand meaning
  • Trained on a single species and population (Atlantic spotted dolphins)
  • Requires extensive labeled data and behavioral context for accuracy
  • Not a commercial product; limited to research collaboration

Best For

Analyzing patterns in dolphin vocalizations (clicks, whistles, burst pulses)Generating realistic dolphin-like sound sequences for interaction studiesSupporting non-invasive field research on dolphin communicationExploring potential for interspecies communication

Alternatives to DolphinGemma

FAQ

What is DolphinGemma?
DolphinGemma is a large language model developed by Google to analyze and generate dolphin vocalizations. It uses the SoundStream tokenizer and is built on the Gemma model family. The model is trained on decades of acoustic data from the Wild Dolphin Project and is optimized for field use on Pixel phones.
Who developed DolphinGemma?
DolphinGemma was developed by Google in collaboration with researchers at Georgia Tech and the Wild Dolphin Project, led by Dr. Denise Herzing.
What type of data does DolphinGemma use?
It uses the Wild Dolphin Project's database of underwater audio and video paired with individual dolphin identities, life histories, and observed behaviors, focused on wild Atlantic spotted dolphins in the Bahamas.