When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.

Microsoft starts rolling out automatic music detection and noise suppression to Teams

The use of online collaboration and communication software has become exceedingly important in this era of hybrid work. One such piece of software is Microsoft Teams, which the company updates on a frequent basis to facilitate the use-cases of its various customers. Now, the Redmond tech firm has announced a bunch of audio improvements for its communication platform, and while not all of them are new, they are certainly welcome.

A Microsoft Teams logo and a music symbol on a colorful background

Something that Microsoft started rolling out recently on Teams is high fidelity music mode. This capability is important because it facilitates numerous use-cases such as music lessons and medical signals during virtual appointments with your healthcare professional. Teams accomplishes this by transferring such audio signals at a 32kHz sampling rate (16kHz bandwidth) at 128kbps. It manages to ensure high fidelity while reducing the bitrate by four times in comparison to lossless encoding. You can find out how to take advantage of this by following Microsoft's guidance here.

A new feature that leverages from high fidelity music mode is machine learning-based noise suppression, that is now enabled by default for Teams customers. This model uses non-speech signals as input and then determines whether it should suppress that noise or not. If it identifies the input as music, such as that being played during a violin lesson, it alerts the user that they should enable high fidelity music mode. If it detects a false positive, the user can simply dismiss that banner. And if the ML model actually detects undesirable noise that is not music, it will automatically suppress it.

Microsoft says that it built this model by training a convolutional neural network on a training set containing one million audio clips with sound and music. The training set included sounds from a variety of environments and instruments to cater to a large variety of use-cases. The output model was then tested on a collection of 1,000 audio clips, which yielded an accuracy of 81%. Microsoft claims that its model outperformed all published research in this area, and you can check out its research paper here.

ML-based noise suppression has now been enabled by default for "most" Teams users, while automatic music detection will be rolled out generally in the coming months.

Report a problem with article
The Vodafone logo on a dark red and black background
Next Article

Vodafone UK will begin 3G switch off from 2023

Previous Article

Twist is an MIT-developed language that attempts to address bugs in Quantum Computing

Join the conversation!

Login or Sign Up to read and post a comment.

6 Comments - Add comment