Scientists build AI tool that could help decipher animal voices

For years, researchers have wondered how animals communicate with each other. Birds sing complex songs, whales chant, and bats communicate through ultrasound conversations. However, deciphering these calls and songs is not always an easy task. Traditional approaches typically try to fit calls and songs into neat boxes, but they don’t always capture the subtle details that are important for understanding language. This is where a new computational model called Chatter comes into play. It was developed as a Python library and uses machine learning and information theory to analyze animal sounds and songs in more detail. The model, first introduced in a 2025 arXiv paper, aims to identify communication patterns across a wide range of animal life.

Rather than forcing sounds into fixed categories, Chatter treats vocal sequences as continuous patterns that change over time. According to the arXiv study, the software maps these signals into what researchers call latent space. Latent space is a mathematical representation that captures subtle relationships between sounds. This method allows scientists to track how vocalizations change, repeat, or differ throughout a sequence. To perform this task, the system requires advanced machine learning techniques. Additionally, certain models, such as variational autoencoders and vision transformers, convert raw audio recordings into compact representations that focus only on the meaningful features in the data. These algorithms allow us to easily measure qualities such as predictability, novelty, and complexity of animal calls without the need for manual labeling (arXiv).

Chatter’s versatility is one of its most notable features. The software has already been tested using records from a large number of animals, including birds, bats, whales and primates, demonstrating that the same analytical framework can be applied to different species groups. As the researchers also point out, taxonomic flexibility is most valuable in comparative studies that examine how communication systems evolve (arXiv). Chatter also includes a complete analysis workflow designed to take researchers from raw recordings to interpretable results. The system handles preprocessing, segmentation, model training, and feature extraction all within a single pipeline. The documentation available from the Chatter project’s official website includes installation instructions, tutorials, and sample datasets to help researchers use the tool effectively.

Chatter provides a more realistic way to investigate animal communication by viewing animal sounds as evolving patterns rather than static boxes. Developers have realized that many natural communication methods feature smooth transitions that were not captured by previous methods. As machine learning technology improves, tools like Chatter may prompt scientists to uncover complex sound patterns found in the animal kingdom that may have long been hidden.