Baidu, the Chinese company that runs the country’s largest search engine, has filed a patent application for an artificial intelligence system intended to translate animal sounds into human language. The filing, published by the China National Intellectual Property Administration in May 2025, was first reported by Reuters on May 8.
What the patent describes
According to the application, the system would gather several kinds of data about an animal: its vocalizations, its behavior and body movements, and physiological signals such as heart rate. Machine learning models would then analyze those inputs together to identify the animal’s emotional state, for example whether it is content, anxious, frightened or in pain, and map that state onto human language so that an owner or handler could understand it.
The filing describes the use of natural language processing alongside pattern recognition on audio and sensor data. Baidu told Reuters that the project was still at the research stage and that there was no timeline for a product. The initial focus appears to be on pets and farm animals, where the practical value of knowing whether an animal is distressed is clearest.
Part of a wider effort
Baidu is not the only large technology company working in this area. In April 2025, Google announced DolphinGemma, a language model trained on decades of recordings from the Wild Dolphin Project to identify patterns in dolphin vocalizations and, eventually, to help researchers test whether two-way communication is possible. Several academic groups and non-profits, including Project CETI’s work on sperm whales, are applying similar methods to other species.
The underlying technical idea is the same across these efforts. Animal sounds are converted into digital representations, the representations are searched for recurring structure, and that structure is correlated with observable context such as behavior or environment. What Baidu’s filing adds is the ambition to fuse audio with movement and physiological data and to produce output in ordinary human sentences rather than researcher-facing labels.
What it would and would not do
Even if the system works as described, it would not let a person hold a conversation with a dog. The output is an inference about emotional state, not a translation of meaning in the linguistic sense. Scientists who study animal communication caution that it is not established that most animals have anything resembling language, and that models trained to label emotions can only be as good as the human annotations used to train them.
There are also practical questions about trust. A device that tells an owner their cat is “happy” invites confidence that the science may not support. A patent filing is a statement of intent and a claim on an idea, not evidence that the idea has been demonstrated. Baidu’s own description of the work as early-stage research is the right frame for now.
Still, the filing signals that interpreting animal signals has moved from a niche research question to something a major consumer technology company is willing to invest in and protect. Whether the first result is a better pet-care gadget or a tool for wildlife conservation, the work will be judged by whether its interpretations hold up against what animals actually do next.
