Real-time dolphin whistle recognition device using Arduino Mega, MAX4466 microphone, and LCD. Trained custom Hidden Markov Models (HMMs) on Sarasota Dolphin Whistle Database to classify vocalizations and predict behaviors (e.g., diving).
Implemented Viterbi decoding on microcontroller with beam search for memory efficiency. Created Python training pipeline for emission/transition probabilities. Final device performed live classification and displayed predicted behavior in real-time.