Chinese AI Pet Translator Claims 95% Accuracy, but Scientists Say Your Dog Is Not Actually Talking to You

A Chinese startup says it has built an AI-powered collar that can translate your pet's barks and meows into human language with 94.6% accuracy, but there is no published science to back that number up.

Hangzhou-based Meng Xiaoyi, which markets the device internationally as PettiChat, opened pre-orders earlier this month and has already secured more than 10,000 reservations at 799 yuan ($118, or roughly £88) per unit. The company has also claimed approximately $1 million (£741,000) in angel funding to support production.

How the Device Supposedly Works

The 27-gram collar clips onto a pet's neck and uses built-in microphones, motion sensors, and accelerometers to capture vocalisations, posture, and body movement. That data is then processed through Alibaba Cloud's Qwen large language model, which the company says was trained on more than one million vocal and behavioural samples across breeds. The device delivers a translated sentence to the owner's smartphone app in roughly 1.2 seconds, according to Meng Xiaoyi's own press materials.

Founder Li Jingyuan has described the underlying technology as an 'Animal Behaviour World Model' that processes visual, auditory, and behavioural signals together. The Kickstarter campaign, which ran through 14 May, listed a $119 (£88.27) 'Super Early Bird' tier and promised retail shipping by the fourth quarter of 2026.

The Missing Evidence

The 94.6% accuracy figure is entirely self-reported. No peer-reviewed study, independent laboratory test, or published dataset supports the claim. The company has not disclosed its testing methodology, sample sizes, or how it defines 'accuracy' in the context of translating animal sounds into human sentences.

That gap matters. Independent research on AI-based animal emotion detection places the accuracy of acoustic signals alone at roughly 57%, with multimodal approaches that combine audio, video, and posture data reaching a ceiling of around 89% to 92% in controlled environments. The leap from classifying a pet's broad emotional state to producing specific human-language sentences is one that no published study has demonstrated.

Animal behaviour researchers have consistently warned that while machine learning can identify general patterns of distress, excitement, or calm in animal vocalisations, the idea of 'translating speech' from species that don't use language the way humans do remains scientifically unsupported.

A Startup Built on Buzz

Meng Xiaoyi was founded in January 2026, giving the company less than five months of existence before launching a consumer product with bold accuracy claims. On Chinese social media, critics have called the device a 'human intelligence test' rather than a genuine translator. HotHardware described the company's demo videos as 'undeniably choreographed' and called the concept of turning a bark into a full sentence 'bordering on science fiction.'

The pet technology market, however, is booming. China alone is projected to have 126 million pets among urban residents, and AI wearables are attracting growing investor interest. The commercial demand for deeper connections with pets is real, even if the science behind this particular product is not proven.

What Consumers Should Know

The viral appeal of talking to your dog is obvious. But when a company founded months ago can claim near-perfect accuracy, collect thousands of pre-orders, and raise $1 million in funding without releasing a single page of verifiable data, it raises questions that go beyond pet communication.

Consumer protection frameworks for AI product claims remain thin in most markets. Until Meng Xiaoyi publishes its methodology and submits to independent testing, the 94.6% figure is a marketing claim, not a scientific one. Your dog may have a lot to say, but this collar hasn't proven it can tell you what that is.