Xiaomi Releases Open-Source Spatial Audio Understanding Model MiDashengLM-Spatial
Xiaomi has launched MiDashengLM-Spatial, an open-source large model designed to understand not just audio content but also its direction and spatial distribution. The model's weights and code are now publicly available.
Chinese technology company Xiaomi has released MiDashengLM-Spatial, a new large language model focused on spatial audio understanding. The model is designed to interpret not only the content of sounds but also their origin and how they are distributed within a space.
Built upon Xiaomi's previous MiDashengLM-7B audio model, the new MiDashengLM-Spatial adds capabilities for spatial perception and inference. Xiaomi states this is the industry's first open-source, end-to-end spatial audio understanding model, advancing the field beyond simple sound recognition to comprehensive spatial context awareness.
The company intends to integrate this new model into its smart home devices and automotive cabins. The goal is to enhance user interaction experiences within its "human-car-home" ecosystem by leveraging the model's spatial audio comprehension capabilities.
The model's weights and inference code have been released under the Apache 2.0 license, allowing for widespread adoption and further development by the research community. This initiative aims to drive progress in audio AI and foster new applications.