Xiaomi has launched a voice of voice. It is not for mobiles, it is for the war of cars

Xiaomi was one of the first brands to announce a Voice assistant with ia For its mobile phones, although it is little known because it only works in China. Seven years later, the Asian giant has announced a new model of voice, but this time its approach is not in mobile phones, but in cars and the connected home.

MIDASHENGLM-7B. Xiaomi has baptized his new model with this name so unattractive and difficult to remember. It is composed of two key parts: the audio encoder with the dasheng and the decoder of ALIBABA QWEN2.5-OMNI-7B. Combined, the system is able to recognize not only our voice, also environmental sounds, music and background noise. Xiaomi presumes that offers a “first level performance in 22 public benchmarks.” Specifically, he has surpassed OpenAI whisper in nonverbal audio understanding tasks.

In the car. Xiaomi has already found 30 applications of its new voice model for different products. Voice control includes the acoustic environment and responds according to the context, for example if there is an unusual sound in the car. They have also thought of a function that helps us improve pronunciation to learn languages while driving and the possibility of “waking” the car with the voice even before we enter the cabin. Its ability to detect anomalous sounds also makes it useful from the point of view of security, both of the car with a more powerful anti -theft mode, and at home through smart speakers.

At home. The new model enables the activation of intelligent functions through sounds, such as that the lights are linked when applauding or the air conditioning goes out when we go out the door, without having to ask with the voice. Xiaomi ensures that its system has a very low latency and a large parallel processing capacity, allowing it to function in devices with few resources, such as cameras or speakers, and maintain good performance in environments with many devices connected as a house.

Open source. China has chosen the Open-Source side in the AI race And the new Xiaomi voice model follows the line. Midashengglm-7b is open source and operates under Apache 2.0 license, which allows commercial use and free modification. That opens the door to its use by other developers and also in the academic field. In addition, Xiaomi has made public all the data that has been used for model training. The objective is clear: attracting the developer community and that its audio ecosystem becomes standard, strengthening its competitive position.

Conversational experience. The automobile industry is taking a turn in which The software is positioned as the differentiating factor number one. It is no longer only if it runs more or how comfortable it is to drive, they are the autonomous driving systems, the interface of its screen and especially voice control. According to This studyvoice attendees with AI will be a standard in cars by 2033. Who has the best conversational experience will have a clear advantage and Xiaomi has taken an important step in this direction.

Cover image | Xataka with Zky.icon icons

In Xataka | Xiaomi continues to lose money with their electric cars … but they are being their greatest success