Google Meet gets AI-powered noise cancellation: Here's how it works
In a bid to take on Zoom and cash in on the exploding demand for video conferencing, Google is updating its premium virtual meeting app, Google Meet, with the much-needed noise cancellation capability. The feature runs with the help of machine learning algorithms and comes handy in improving the quality of group meetings. Here's all you need to know about it.
Filtering out all the background noises
Rolling out to G Suite users in a phased manner, noise cancellation is an audio setting aimed at reducing ambient noise in virtual meetings. The feature works automatically and filters out all unnecessary background noises - starting from sounds of traffic, dog barking, and baby crying to those made by pen clicking and wrapper crinkling - while keeping the main voice of the speaker.
How to use the feature?
When Google announced noise cancellation for Meet, it confirmed that the feature would first come to G Suite Enterprise and G Suite Enterprise for Education customers. Now, as the roll-out continues, it is becoming accessible for Meet users on the web. It is activated by default but can be turned off by heading over to audio settings and turning the noise cancellation toggle off.
Audio is processed on Google Cloud
Speaking to VentureBeat, Serge Lachapelle, G Suite's Director of Product Management, said the capability works with the help of machine learning algorithms. Basically, the audio of the speaker is sent securely to the cloud where a machine learning algorithm analyzes it to detect the main voice and remove all the ambient noises surrounding it in real time.
Quality degrades a bit with processing
Once the processing takes place, the filtered out voice - free from ambient noise - is delivered to the person on the other end of the call. Overall, the quality of the voice comes out pretty clear, but the segments where the processing has been done sound a little degraded or low. Notably, Lachapelle claims that it will improve with time.
External meetings will not be used for training
The feature has been trained on noise and clean speech fetched from the company's internal meetings and YouTube videos over the last year. It will be improved with more data but Lachapelle emphasized that the data will not be taken from external meetings on Meet.
