Soon, each video depicts Chwehik and you're talking, even without the need for voice, too would make the artificial intelligence of Google recognizes the content of your words
A team of researchers from Oxford University and Deepmind subsidiary of Google, specializing in the field of artificial intelligence (AI), created an application for the analysis of lip movements and extract the content of speech, where their efforts crowned with great success since the program was much more efficient than humans.
The team trained neural their network (or neuron) over 5000 hours of television programs of the BBC, and the sample contained a clause 118000 17500 and the word unique.
As a result, the two teams achieved a pass rate of 46.8%, the proportion of very successful even though they look weak ratio, it must be noted that the research team had to bring specialists in thefield and offered them the same sections offered on artificial intelligence inadvertently speech recognition content than just Hrcah control the lips, the result was 12.4% only .
The team explains in research Alorguet published on arXiv that the biggest difficulties facing the reading and analysis techniques come from the lips of common words in pronunciation but different meanings, it is difficult to predict in this case to be in the proper sense of that word.
This new technology for intelligent artificial expected to have uses and applications of a wide, among them the possibility of water a message or give a specific command of Assistant personal phone even though you are in a very noisy environment, the revival of silent film from the archive, improving the level of voice recognition and other useful applications.
However, there is a secret that technology like this will be something joyous for those working in the field of espionage, as it will open new horizons and possibilities for an unprecedented spy and track the speech content without the need for a microphone close to the person to be targeted.
While the research team emphasizes that their program needs a high-resolution videos for speech recognition content and Cameras surveillance, for example, are not valid at all for this purpose, but the artificial intelligence in the continued progress in this field, given the attempts of other research teams in order to reach better results.
0 Comments