Parasite number one when it comes to video chat, background noise will be reduced in the Microsoft Teams application. How? Through machine learning and the use of a million audio samples to try to eliminate it.
Microsoft has developed several new audio features for Teams. Machine learning-based noise suppression can automatically detect background noise that needs to be removed, allowing the communications app to make speech clearer in meetings. An automatic music detection feature is also on its way to Teams, but it will take a few months to try it out.
The high-precision music mode transmits audio at a sampling rate of 32 kHz (16 kHz bandwidth) at 128 kbps. The mode reduces the bit rate by a factor of four compared to lossless encoding. Microsoft created the new feature with music classes and performances in mind. The company says that the high-fidelity music mode can also clearly transmit medical signals during virtual reception.
Non-vocal sounds automatically removed
Microsoft provides audio samples of music from a normal appointment and music in high-fidelity mode for comparison. The normal appointment music sounds a bit hollow; it is clearly transmitted over the Internet and compressed. In contrast, the audio in high-fidelity music mode sounds richer.
Microsoft also has news about another audio-related feature, machine learning noise reduction. Most Teams users should now have this feature enabled by default. It determines which non-speech sounds should be removed, making it easier to hear what people are saying during a meeting. Teams will automatically determine if a non-speech sound is music and display a prompt to turn on high-fidelity music mode.
More than a million audio clips were used to train Microsoft's deep neural network. To test the feature, the company invited musicians playing several instruments in different genres.
Source: Windows Central