Scientists at Oxford University has revolutionized the AI surveillance and built a machine that can lip-read better than human beings with funding help from Google’s and able of facial recognition to lip reading.
Oxford University students showed their skills by developing the machine that works on artificial intelligence and can read the lip with a higher level of accuracy than even the highest trained experts and it can actually read lips 93.4 percent of the time.
The artificial intelligence-based system – the LipNet – it is the machine that watches the video of the person speaking and it will determine from its lips and matches the text with 93 percent of accuracy – the scientists said.
The experts also said that the system is needed to be tested in real life situations – as lip reading is the tricky process to find out the words of the person from its lips movement.
The experts say that the purpose of the machine is to improved hearing aids, silent dictation in public places whereas speech recognition and biometric identification. The computer is trained with three-second videos of different people of more than thirty thousand, through which machine is capable of learning different lip movements with the word being spoken.
The system is not revealed yet in the world but the videos are launched by the oxford experts and it showed that human testers have an average artsandhealth.ie/finasteride/ error rate of 47 percent but the fact that when this machine is tested it is only 6.6 percent was wrong – the researchers tested this machine against three experts.
Lip Net is also able to make great changes in the surveillance system i.e. CCTV cameras – the experts said there is no software built for the purpose but this machine can recognize the words if the video is captured via high resolution. This may help in recognition of the voices of persons speaking in video or you can speak the silent movie with this machine.
The artificial intelligence of lip net is fed by the experts – stores videos labeled with correct text and follows a grammatical pattern but has a limited vocabulary.
When this device is able to a recognition of CCTV footage than – it will be useful for fraud protection and its pure poison from civil liberties – the machine works in a set of command, color, preposition, digit and verb with thirty-four speakers that speak one thousand sentences each.
The experts said that the machine needs a lot of vocabulary for full verification whereas it needs more work on different face shapes – as well as more people with different accents – if you are worried about your recognition of words than wear a mask against surveillance cameras to save your conversation.