MiDashengLM-Spatial: xiaomi's new spatial audio understanding model

Xiaomi has unveiled MiDashengLM-Spatial, an open-source spatial audio understanding model, as part of its Dasheng series. The model is fully open, with weights and inference code made public under the Apache 2.0 license. It builds on the MiDashengLM-7B model, adding spatial perception capabilities to enable the model to understand not only what is being said but also where the sound is coming from. Xiaomi claims it is the first end-to-end spatial audio understanding model to be open-sourced. The model is expected to enhance user experience in Xiaomi's smart home and car cockpit scenarios. The project's website, GitHub repository, and HuggingFace page provide access to the model's structure and weights.