GOOGLE IS looking for people in the crowd again, but this time, it's not faces, it's voices.
The new Google tech uses facial recognition to find voices in, say a crowded pub, to boost the sound wave over the others and make it audible.
By teaching a neural network what different voices sound like when it's quiet, it was able to pick them out from part of a crowd. You can even filter the background noise and ‘babble'.
"We believe this capability can have a wide range of applications, from speech enhancement and recognition in videos, through video conferencing, to improved hearing aids, especially in situations where there are multiple people speaking," explains Google.
In fact, Google is looking to add it into Duo and Allo.
"A unique aspect of our technique is in combining both the auditory and visual signals of an input video to separate the speech. Intuitively, movements of a person's mouth, for example, should correlate with the sounds produced as that person is speaking, which in turn can help identify which parts of the audio correspond to that person," explains Google in a blog.
"The visual signal not only improves the speech separation quality significantly in cases of mixed speech (compared to speech separation using audio alone, as we demonstrate in our paper), but, importantly, it also associates the separated, clean speech tracks with the visible speakers in the video."
Thing is, therefore, if it can be used for this, it has a lot of other applications that are a bit more… well… Black Mirror-y.
If you're on camera, and your voice is trained within the neural network, how do you know you're not being spied on?
And yes, sure, it might make Google Assistant work better when the TV is turned up loudly, and it might make it easier to transcribe an interview in a noisy environment, but more worryingly, it can't do anything but impair your privacy.
This is just how HAL found out he was going to be switched off in 2001: A Space Odyssey and thus went batshit. µ
We doubt Big Blue would take payments in voucher form
But the service still has 130 million members, so it's not exactly in trouble
Zhang could be jailed for up to 10 years if found guilty
America seethes as offers are replaced by error pages. With dogs