Google Artificial Intelligence is capable of posing as a human being • HWzone
ComputersTechnology

Google's Artificial Intelligence can impersonate a human being

In laboratories DeepMind was developed Which allows computers to talk to us in a voice that sounds much more human and natural

Get updates from everyone in TelgramGet updates from us all at TelgramJoin the channel now


DeepMind It is a British company established in 2010 for the purpose of researching and developing artificial intelligence. Initially, the company developed capabilities that allowed the computer to learn to play old computer games such as "Space Invaders" by analyzing only pixels as input to the computer. After a great number of games and self-learning, the computer has developed the ability to win this game and other 48 other computer games. In 2014 the company was bought by .

This year, the company came back to the public agenda after developing a program called AlphaGo, which managed to win a world champion in the GO game. The company is trying to create an artificial intelligence capable of solving complex problems without having to teach the computer how to do so. According to the company, if the developers succeed in their goal, they will be able to solve important problems in medicine, climate and the study of the universe. Among other things, artificial intelligence helped reduce At approximately 40 percent The cooling costs of the server 's huge server .

DeepMind has now released a new technology called WaveNet, Which makes it possible to create a computerized sound that will sound more natural than ever before. The way human voice is produced today in computer assistants such as או It is by using many human speaker sound files. The same sounds are cut and organized by the computer to create a relatively high-level computerized voice - but with great difficulty in creating the appropriate tone, emotion and intonation when assembling a complete sentence. When we hear a voice produced by a personal computer assistant, we experience a human-sounding voice, yet with a rather robotic feel. WaveNet, too, compiled the cut and arranged sound samples into an entire sentence and added human intonation and nuance to it, drawing on additional human sounds from which the system learned.

So while a computerized voice today may sound like this:

The sound produced by WaveNet is closer to a human voice and sounds like this:

The company surveyed hundreds of people to determine the grade of humanity from 1 to 5. The score for the new technology in English was 4.21 compared to the existing alternative, which received an 3.86 score. Incidentally, even a human voice was not accepted by the 5 scorers. The survey's score was 4.55, so WaveNet received a score very close to the human level.

Human Voice Survey of WaveNet System
Human Voice Survey of WaveNet System

Today's personal computer assistants such as Siri and Cortana are gaining momentum as technology advances. But their robotic expression creates in many of their human users a sense of remoteness known in the world of robotics and artificial intelligence as "Valley of the Strange", A feeling that is created when something looks or sounds human but is not quite so. WaveNet can certainly help solve the alienating robotic sensation created by the familiar computer voice today.

At this point WaveNet is not suitable for integration with home computers and smartphones because of the consumption of too many computing resources to produce natural voice. Even with this technological breakthrough, it seems that we are still very far from a product that will sound like the personal assistant Samantha in her science fiction film. But as time goes on, improving the power of computing and the efficiency of WaveNet technology, we may sound more friendly and natural than ever before. However, there is another step in this direction.


Tags

3 תגובות

Leave a Reply

Back to top button
Close
Close