Google has demonstrated the AI noise canceling feature in Google Meet. Demo on YouTub reviewed by G Suite Chief Operating Officer Serge LaChapelle. The firm previously promised to first implement sound cancellation for G Suite Enterprise and G Suite Enterprise for Education users.
With the help of a new sound suppression tool, the recognition of ordinary sounds arising during a conversation is carried out: the crunch of a bag of nuts, the click of a pen, the clink of glass. the function removes such noise from the conversation, filling it with complete silence. She will only focus on her voice. Even with poor speech audibility, excess noise will be eliminated.
The function works in the cloud, it does not need to be downloaded and does not require special hardware. See how it works in the view:
It is worth noting about the presence of machine learning technology. According to the firm’s promise, this tool will be improved.
According to Serge LaChapelle, sound cancellation was used internally by Google in 2019.
Active development began after the company was acquired by Limes Audio in 2017. A prototype was built the following year. Initially, the team used their collected records to train the version.
LaChapelle says the company uses a soft approach and is not going to seek the ability to suppress all extraneous sounds.
In his opinion, the decision to use a controller to switch the sound suppression mode appeared after the idea was demonstrated to the company’s team. At that time there was applause, but it was suppressed by the function.
According to LaChapelle, the tests show successful suppression of sounds produced by household appliances and musical instruments, especially drummers. Using the function, you can extinguish music in the background, but not interrupt the singing of people. Different languages were used for testing.
A separate difficulty was the presence of noisy equipment near the microphone, for example, a keyboard. LaChapelle placed the caller iPhone next to the microphone. The ringtone sound was not loud, but the speaker’s speech was not overlapped. In addition, LaChapelle’s voice became quieter with a slight distortion.
“The voices of Google Meet attendees are recorded using a Google datacenter, where they will be passed through a TPU machine learning model, re-encrypted, and sent back. With sound cancellation, a device that removes noise with a key shared by all participants reads the information, then the sounds are removed, then information is sent back using the same key “- according to Lashapelle.
Meet audio canceling will be turned on by default for G Suite business customers this month. It can be disabled in the settings.
The cloud makes it possible to enable sound cancellation on the server side without changing the quality, regardless of the ability of the client’s device directly. Then the code needs to be optimized.
LaChapelle says it takes a lot of money to add an extra stage of speech processing to all meeting attendees on Google Cloud.
For this reason, the feature is provided initially to paying G Suite customers. Then a more global deployment is supposed.
Read other news in the section Hi-tech News.