Meet Google’s New Virtual Assistant That Speaks Like A Human

Artificial intelligence is increasingly getting developed and with better capabilities to interact with humans.

During Google I/O, Sundar Pichai introduced a new component of his wizard, Google Duplex, a feature that allows artificial intelligence to make calls.

In recent times, artificial intelligence has been one of the areas most worked, leading to major evolutions. If virtual assistants are where this technology stands out, the truth is that today is present in more and more components, as is the case of machine learning and the cameras of the latest smartphones.

During Google I/O, the American giant has shown yet another innovation in this area, now offering the possibility to its artificial intelligence to make calls as if it were a human being.

Google’s new Artificial intelligence that speaks like a human

The tech giant Google is one of the brands that has its most bet on the development of artificial intelligence and the new Google Duplex is proof of this.

During Google, I/O, 35 minutes after starting his presentation, Sunda Pichai, CEO of the tech giant Google, surprised everyone present by asking the virtual assistant to mark a haircut and it has responded to the letter.

Then there was the demonstration of the new Google Duplex, with artificial intelligence making a call to a hairdresser and interacting as if a human being were trying to mark the haircut. Then a second example was presented, this time in a restaurant reservation, where Google Assistant once again performed remarkably.

During the call, Google Duplex allowed the artificial intelligence to simulate imperceptibly human conversation, with some details such as “Hum …” and “mmm-hmm” being present.

But what is Google Duplex and how does it work?

Google Duplex, according to the tech giant Google, is a conversational system composed of a recurrent neural network (RNN) developed based on Tensor Flow Extended. In addition, it makes use of speech recognition technology (ASR), in addition to analyzing various parameters understands what the speaker is saying and analyzing the context. To control intonation, Text to Speech (TTS) based on Tacotron and WaveNet was also used, and some “Speech Dysfluencies” (Hum ..; Mmm-Hmmm) were introduced in order to humanize speech.

o get to this level, the tech giant Google has trained its system with anonymous calls and in various contexts. Thus, the virtual assistant is able to maintain autonomous conversations without human intervention, although there is a system to alert a human operator if the task is not completed.

At the moment, Google Duplex is still limited to reservation systems, helping to limit the number of scenarios and issues raised.

Scary or fantastic?
Although this is a well-applauded evolution, several people have also protested against this evolution, even calling it frightening.

Given that the caller hardly realizes that he is talking to a robot, they consider it important that a system is created that identifies when these calls are made.


The truth is that the tech giant Google has taken a giant step in its artificial intelligence, but it is something that the American giant has to think how to control so that it is not used with less sensible intentions.