Quick summary
Xiaomi has released an open-source AI voice model offering advanced sound recognition properties.
The MiDashengLM-7B model can interpret a wide range of audio inputs and is already used in Xiaomi's cars and smart home devices.
There's no escaping AI at the moment, with fierce competition to develop and deploy AI across a full range of applications. Xiaomi, hot off the back of shaking up the world of electric vehicles, has released a new AI model.
The new MiDashengLM-7B is an open-source AI voice model, meaning it's available for other businesses to deploy. It's based on Xiaomi's AI tech that, according to ITHome, can understand speech, environmental sounds and music – and has already been put to use in smart home devices and cars from the company.
It combines two elements, the Dasheng audio encoder from Xiaomi, alongside the Qwen2.5-Omni decoder from Alibaba.
In the real world, this is the AI tech that sits behind the enhanced sentry mode offered by the new Xiaomi YU7, with Xiaomi saying that it already powers more than 30 AI features that the company offers.
Xiaomi YU7
It's no slouch either, with breakthroughs in 22 public evaluation tests. Xiaomi also declares all the sources that this model is trained on, which is said to be 100% public data.
This language model is designed to understand what's happening around the user with that full spectrum of recognition of sounds, to dive a little deeper and improve responses through the additional context it can interpret.
In the smart home, this model can provide monitoring for specific sound triggers, either as an alert function or to provide gesture controls based on sounds. The low latency and efficiency of the model mean it can trigger faster than rivals, making it a good candidate for sentry modes in a car.
While Xiaomi has used this model effectively within its own devices, making it open source will mean that others will be able to take this model and use it themselves. That might garner the interest of others developing within the automotive or smart home space.
The open-source nature of this model will present a challenge for major tech companies pushing their own licensed AI platforms, like OpenAI or Anthropic, with developers able to get access to the model easily. That could help accelerate the evolution of new AI features and applications.
Xiaomi has said that it continues to develop the efficiency of the model, looking to offer offline access and enhanced features like sound editing.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.