Xiaomi open-sources MiDashengLM-Spatial spatial audio model

Xiaomi released MiDashengLM-Spatial, an open-source ~8B-parameter audio model that identifies not only what a sound is but also its direction, distance and movement. Weights and inference code are published under Apache 2.0, building on MiDashengLM-7B with a Spatial-Dasheng encoder.

• ~8B parameters, Apache 2.0 license for research and commercial use
• Spatial cues require binaural input; mono audio runs without direction
• Xiaomi calls it the first open-source end-to-end spatial audio model
• Dasheng tech already powers 30+ smart home and car cockpit apps