Russian AI Can Now Read Emotions More Accurately Than Ever

A team from Russia’s Sberbank and the National Research University Higher School of Economics has developed an advanced AI model capable of interpreting human emotions by analyzing facial expressions, vocal tone, and speech patterns at the same time, according to CNews.

The technology merges temporal convolutional networks (TCNs) with transformer-based multimodal architectures, enabling dynamic emotional state recognition over time. It outperforms previous models—those limited to facial analysis—by more than 10% in accuracy. This makes it applicable in business environments, public safety systems, and social services.

According to Andrey Savchenko, head of Sberbank’s Center for Applied AI, the new model makes human-AI interaction more intuitive and meaningful. It’s especially promising for industries where emotional insight can enhance user experience. The system has already excelled in early testing and is being considered for integration into virtual assistants, security platforms, and even telemedicine. Importantly, the model functions even when the user’s face is obscured or voice is faint.

Researchers also see use cases for the technology in call centers, product testing, and real-time behavioral monitoring—like detecting aggression or panic in sensitive scenarios. Further improvements to the model are already underway, and international interest is growing.