Xiaomi has opened the source code of a powerful voice and model Midashenglm-7b, designed to control cars and smart houses.
The model is available on the Huging Face platform with demos and scales, is distributed under the Apache 2.0 license and already demonstrates excellent results in 22 tests of multimodal models, overtaking even Whisper in tasks outside of speech recognition.
Midashenglm-7b combines voice control, analysis of surrounding sounds and music, supports 50+ languages and is able to recognize individual users without activation words.
Equipped with semantic mapping, it analyzes the emotions and spatial characteristics of sound, which makes it universal for real -time use in household appliances, electric vehicles and training systems.
The discovery of the code emphasizes the Xiaomi strategy-to develop the Ecosystem “Man-Domobile” and compete with Western AI giants through openness.
