Long term work and Machine Learning Approaches to DolphinGemma
summary
Four decades of annotated underwater video and high‑frequency audio from Bahamas Atlantic spotted dolphins form the training set and permit correlating acoustic motifs with behavior; unsupervised clustering and grammar induction recover signature whistle sequences and other predictive motifs. A working underwater wearable acoustic keyboard (Pixel 9 + UMIK mic + surface exciter speaker + bone‑conduction cueing) reliably plays and recognizes prompts, and a DolphinGemma generative AI produces short (≈3 s) synthetic echoes used to augment training; surprising result: VCM3 “negative‑space” mimics comprise >50% of responses.
AI-written summary · about
chapters (6)
- 0:00Welcome to dolphin research
- 1:24Dolphin data collection methods
- 12:54Dolphin sound and behavior analysis
- 22:56Dolphin communication with tech
- 34:18DolphinGemma model and uses
- 47:50Q&A
related talks
57mDecoding Dolphin Communication
01h 01mIndividual Discrimination of Bottlenose Dolphin Signature Whistles Using Deep Learning
01h 03mGenerative AI and What Is Meaningful Sperm Whale Communication
53mAI as a tool for discovery in animal communication
56mWhistle variability and social acoustic interactions in bottlenose dolphins
25mThe Sound of Life
01h 04mGPT-Whale Language Models at Tools for Structure Discovery and Hypothesis Generation
29mBioacoustics and Machine Learning as Key Tools in Conservation