Xiaomi Company: the international performance of voice recognition algorithm ranks first, and the audio marking task has made important progress.

Thanks to CTOnews.com netizens for the delivery of clues such as Rain and Snow, StarDevOps, vegetable Dog and wuzhuce! CTOnews.com September 13 news, Xiaomi company today released news, Xiaomi self-developed voice recognition algorithm in the audio marking (Audio Tagging) task has made important progress.

According to reports, the audio marking model, which uses the audio data of the open data set AudioSet-2M as the training set, has exceeded the score of 50 mAP for the first time. This breakthrough indicates that the Xiaomi voice recognition algorithm has ranked first in the international performance.

Xiaomi said: "in the future, we will continue to explore new heights of technology to bring users a more efficient and accurate voice recognition experience in the use of mobile phones, speakers, bracelets, CyberDog and other rich devices."

As previously reported by CTOnews.com, Xiaomi is also making efforts in the AI model, and combined with Xiao Ai to further expand user coverage. Combined with the progress of voice recognition algorithm, Xiao Ai is expected to usher in a better experience.

In his previous annual speech, Lei Jun revealed that Xiaomi has been laying out the AI since July 2016 and set up a large model team with more than 3000 people in April this year. The latest 1.3 billion-parameter model of Xiaomi AI model has successfully run on mobile phones locally, and some scenarios are comparable to the results of 6 billion-parameter model running in the cloud.