Project CETI researchers used custom AI models to analyze sperm whale codas, identifying vowel-like patterns described as one of the closest parallels to human phonology found in animal communication.
Machine learning reveals new structure in sperm whale calls, the closest parallel yet to human phonology
Project CETI, the Cetacean Translation Initiative, is using machine learning and robotics to study how sperm whales communicate, and recent peer-reviewed work suggests their click sequences are more structured than scientists once believed. The nonprofit records the rhythmic bursts of clicks known as codas and feeds them into custom-built models that search for patterns a human ear would miss.
A study published in the journal Proceedings B in 2026 concluded that sperm whale coda vocalizations are highly complex and represent one of the closest parallels to human phonology found in any analyzed animal communication system. Researchers documented vowel-like and diphthong-like patterns, suggesting the calls carry layered structure rather than simple repetition. Earlier in the program, the team published an automatic coda detector in Scientific Reports in April 2025 and, in December 2025, presented WhAM, described as the first transformer-based model able to generate synthetic sperm whale codas from audio prompts, at the NeurIPS conference.
“The nonprofit records the rhythmic bursts of clicks known as codas and feeds them into custom-built models that search for patterns a human ear would miss.”
The project is collaborative, drawing on partners including Harvard University for autonomous tracking methods and NYU School of Law's More-Than-Human Life program for the legal and ethical questions the work raises. Decoding animal communication could reshape conservation and even debates over animal rights.
Caution is essential here. Finding structure that resembles human phonology is not the same as translating meaning, and researchers have been careful not to claim they have decoded a whale language. The work depends on large, carefully collected acoustic datasets, and AI pattern-finding can mislead if not rigorously validated against behavior. What the studies do offer is a disciplined, transparent way to ask whether other minds organize sound the way ours do. For a creature humans hunted to the edge of collapse, simply listening this closely is its own quiet act of respect.
How did this story make you feel?
📎 Cite this article
Good News Good Vibes. (2025, December 1). Machine learning reveals new structure in sperm whale calls, the closest parallel yet to human phonology. Retrieved from https://goodnewsgoodvibes.com/en/article/project-ceti-machine-learning-sperm-whale-codas-phonology-2025
https://goodnewsgoodvibes.com/en/article/project-ceti-machine-learning-sperm-whale-codas-phonology-2025
Editorial Team
Our editorial team curates and verifies positive news from credible sources worldwide.
Last reviewed: December 1, 2025
Trending
Papua New Guinea Creates a UK-Sized Ocean Sanctuary in the Heart of the Coral Triangle
Animals · 5 minOpenAI's o1 Reasoning Model Outperformed Doctors at Diagnosis in a Real-World Harvard-Stanford Study
Artificial Intelligence · 5 minTropical Rainforest Loss Dropped 36% in 2025, Driven by a Sharp Reduction in Brazil
Environment · 5 min80-Year-Old Vietnam Veteran William Alvarez Crosses Finish Line in His Fourth Boston Marathon
Sports · 5 minYuvelis Morales Blanco, 24, Wins 2026 Goldman Environmental Prize for Helping Halt Fracking in Colombia
Human Stories · 5 min