Alibaba Launches New AI Model Capable of Recognizing Human Emotions

Alibaba has unveiled a new artificial intelligence model, R1-Omni, capable of inferring human emotions by analyzing video, audio, and images. Developed by researchers at Tongyi Lab, led by Xiaoxing Gao, the model is based on the open-source HumanOmni framework and represents a significant advancement in emotional AI.

The model utilizes Reinforcement Learning with Verifiable Rewards (RLVR) technology, enabling it to understand human emotions with high accuracy by analyzing facial expressions, clothing, and surrounding environments.

It was trained on extensive datasets, including the Explainable Multimodal Emotion Reasoning (EMER) dataset, which focuses on emotion derivation, and the HumanOmni dataset, which contains manually labeled emotional data.

R1-Omni improves three key aspects of emotion recognition: interpretation, accuracy, and reliability, delivering more precise and trustworthy results compared to previous models. This makes it a powerful tool for applications such as customer service, smart vehicles, and more.

Through this launch, Alibaba aims to strengthen its position in the AI sector. The company is currently developing new AI tools and applications across various fields, having already compared its Qwen model with DeepSeek and partnered with Apple to provide AI features on iPhone devices in China.

The R1-Omni model is available for free download on the Hugging Face platform, allowing developers and researchers to further explore and enhance its capabilities. This release is part of global efforts to achieve emotional intelligence, enabling computers to recognize and respond to human emotions effectively.

Such technologies are already being used in diverse applications, from improving customer service through chatbots to monitoring driver fatigue in smart vehicles.