Xiaomi's artificial intelligence voice technology may arrive sooner than expected, overtaking the likes of Alexa and Gemini.
In a groundbreaking move, Chinese tech giant Xiaomi has announced the release of its open-source AI voice model, MiDashengLM-7B. This 7-billion-parameter voice AI model is poised to redefine the boundaries of technology in both the automotive and smart home sectors.
In the automotive industry, MiDashengLM-7B powers Xiaomi’s electric vehicle models such as the SU7 and YU7. The model offers enhanced in-car voice control, enabling a seamless, hands-free driving experience. It also provides real-time pronunciation feedback, aiding language learners during travel. Moreover, the AI model offers 24/7 ambient sound monitoring, detecting unusual noises or potential threats around the vehicle for enhanced security.
The smart home industry stands to benefit significantly from the introduction of MiDashengLM-7B. The model enhances wake-up systems that respond reliably to voice commands, improves gesture-based controls for more natural, contactless interactions, and provides continuous sound monitoring for identifying abnormal events, thereby enhancing home safety and automation responsiveness.
MiDashengLM-7B boasts high computational efficiency, capable of handling large batch processing (512 batch sizes on an 80GB GPU), making it scalable for wide deployment across Xiaomi’s IoT ecosystem, which already connects nearly 944 million devices worldwide. This scalability supports Xiaomi’s strategy of creating a seamless, interconnected user experience that spans smartphones, smart homes, and electric vehicles, differentiating its platform from competitors.
By open-sourcing this AI model, Xiaomi aims to foster innovation and broader adoption in AIoT applications, enhancing user convenience, security, and ecosystem stickiness in connected living and transportation environments. The model, which has already been used in Xiaomi’s cars and smart home devices, has performed well in 22 public evaluation tests.
The open-source nature of MiDashengLM-7B could potentially challenge major tech companies pushing their own licensed AI platforms. ITHome reported that Xiaomi's AI tech can understand speech, environmental sounds, and music. The model, based on Xiaomi's AI tech, combines the Dasheng audio encoder from Xiaomi with the Qwen2.5-Omni decoder from Alibaba.
Xiaomi plans to develop the efficiency of the MiDashengLM-7B model further, with a focus on offline access and enhanced sound editing features. The evolution of new AI features and applications might be accelerated due to the easy access to the MiDashengLM-7B model provided by Xiaomi. This open-source model is now available for other businesses to deploy, particularly in the automotive and smart home industries.
The smart home industry will leverage MiDashengLM-7B for improved wake-up systems, gesture controls, and sound monitoring, making homes safer and more automated. In the automotive industry, MiDashengLM-7B will provide enhanced in-car voice control, real-time pronunciation feedback, 24/7 ambient sound monitoring, and potential threat detection, thus offering a safer and smarter driving experience.
