IT Home On October 16, Xiaomi announced an important new member of the Dasheng series — MiDashengLM-Spatial. The complete public model weights and inference code have been released under the Apache 2.0 license.
In the past two years, Xiaomi has released and made open-sourced the sound base model Xiaomi Dasheng and the sound understanding large model MiDashengLM-7B. These technologies have been implemented in over 30 practical applications in Xiaomi smart homes and car interiors, enabling unified perception and understanding of speech, ambient sounds, and music.
According to the introduction, the new model is based on MiDashengLM-7B as a foundation. It further adds “a pair of ears” to the large voice understanding model, enabling it to not only understand the content of sounds, but also perceive where the sounds come from and how they are distributed in space.
Xiaomi said that MiDashengLM-Spatial is the first open-source end-to-end spatial audio understanding large model in the industry. While retaining basic audio understanding capabilities, it also has spatial perception and reasoning abilities, pushing audio large models from simply “understanding sounds” to “understanding the space where the sound is located”.
This model will be gradually applied in scenarios such as Xiaomi smart homes and car interiors, thereby further enhancing the interactive experience of the “person-car-home entire ecosystem” through spatial auditory capabilities. The relevant test data are as follows:
IT Home provides relevant links:
Project homepage: https://jinbo-hu.github.io/midashenglm-spatial-demo/
Technical report: https://arxiv.org/abs/2610.11156
Evaluation code (GitHub): https://github.com/xiaomi-research/midashenglm-spatial
Model structure and weights (HuggingFace): https://huggingface.co/mispeech/midashenglm-spatial
Advertising statement: The text contains external link references (including but not limited to hyperlinks, QR codes, passwords, etc.) designed to provide more information, save selection time, and the results are for reference only. Articles on IT Home that include external links comply with this statement.