Microsoft launches the new AI-based noise suppression for Teams. This Functionality is supported in Windows desktop clients but the company confirmed that it will come to Mac and mobile devices.
In the blog post, Microsoft gave an overview of how the real-time noise suppression feature uses machine learning to remove noise in Teams meetings. Microsoft created a data set to train the ML model to learn the difference between background noise and clean speech.
The noise suppression feature works by analyzing an audio feed and uses a specially trained neural network to filter out the noise and retain clean speech.
Traditional noise suppression algorithms can adjust simple stationary noise sources like fan noise but an AI-based approach can learn the difference between speech and noise and is able to suppress various non-stationary noise like keyboard typing or food wrapping crunching.
Microsoft said it did not collect data for creating this data set. The engineer extracted specific scenarios from publicly available data and used crowdsourcing techniques for gathering data. They optimized the deep learning model for human perception to run the feature efficiently on all Windows devices.
Microsoft open-sourced the perceptual quality crowdsourcing framework and the data set. In research, the company is inviting developers to improve its ML model. The team hosted two competitions so far.