Xiaomi has officially released MiDashengLM-Spatial, a new addition to its Dasheng series of audio models. The company has made the model weights and inference code publicly available under the Apache 2.0 license, allowing for open access to the technology.
The new model is built upon the existing MiDashengLM-7B architecture. While retaining core audio understanding capabilities, it introduces spatial perception, enabling the system to identify the origin and distribution of sounds within an environment.
Xiaomi describes this as the industry's first open-source end-to-end spatial audio understanding model. It aims to transition audio AI from simple content recognition to a deeper understanding of the spatial context of sounds.
The technology is slated for integration into Xiaomi's smart home devices and automotive cockpits. These applications are designed to enhance the interactive experience across the company's ecosystem by leveraging advanced spatial hearing capabilities.
Over the past two years, Xiaomi has developed and open-sourced the Xiaomi Dasheng foundation model and the MiDashengLM-7B. These technologies have already been implemented in more than 30 practical use cases for voice, music, and environmental sound analysis.
