Home » Technology » Microsoft » Microsoft Teams uses AI to improve speech quality, acoustics and more

Microsoft Teams uses AI to improve speech quality, acoustics and more

Microsoft’s communication solution Teams is increasingly becoming the flagship of Office and the entire company. And so it should come as no surprise that the Redmond teams want to improve with the latest technologies, currently, the language in calls is on. The two-year pandemic has made teams and video conferencing the norm, and many employees have long used such software as a matter of course at home or in the office.

However, the pandemic has also taught us that not everyone has the same (good) equipment. That means: the differences in the quality of video and audio are sometimes enormous. Microsoft currently mainly wants to improve speech quality and has deployed its own AI solutions for this. As The Verge reports, the Redmond-based company is fundamentally overhauling audio capture and transmission. Especially those users who are in rooms with poor acoustics should benefit from this. How Microsoft Teams improves audio quality using AI.

How it all works is also demonstrated in a video: Two of the three improvements are explained relatively quickly and are immediately understandable. This is how the AI ​​or machine learning solves acoustic problems: On the one hand, team members sitting in “caves” are corrected, and the AI ​​eliminates the reverberation that is present in some rooms. On the other hand, there is classic noise reduction, which is demonstrated in the video with a partner operating a loud coffee grinder in the background – this noise is simply eliminated by the team.

Interrupt the opponent

To achieve this, Microsoft analyzed approximately 30,000 hours of voice recordings, using thousands of different devices and simulating approximately 100,000 rooms. Finally, a function has also been implemented that aims to make interrupting conversations easier or better: in team conversations, you can interrupt each other without creating annoying overlaps where you cannot hear the person you are talking to because of the echo.