On January 22, 2026, Lei Jun, the Founder, Chairman, and CEO of Xiaomi, made an announcement that several of Xiaomi's AI innovations have been hand - picked for presentation at the highly prestigious international academic conference, ICASSP 2026. These selected accomplishments cover a wide range of AI areas, such as audio comprehension, evaluation of music generation, general audio - text pre - training, and video - to - audio synthesis.

Among these innovations, the ACAVCaps dataset stands out. It allows for a detailed description of audio content through an automated annotation system. This dataset consists of roughly 4.7 million audio - text pairs. The FedDCG framework is also noteworthy. For the first time, it tackles both category and domain generalization problems simultaneously within a federated learning environment.

The FUSEMOS architecture employs a dual - encoder approach to boost the accuracy of perceptual evaluation in music generation. And the GLAP model successfully achieves cross - audio domain and cross - lingual alignment between audio and text. It can even support keyword recognition in as many as 50 languages.

ICASSP 2026 is set to take place in Barcelona, Spain, in May 2026. It is widely recognized as one of the most authoritative and influential academic conferences in the global audio sector.