Google duplex almost a human

By  Pooja Chaudhary    17 - 01 January, 70
google-duplex-almost-a-human

Share article on social media


Imagine yourself rushing to work with a file in one hand and a coffee mug in another and then suddenly you realize that you have a meet and greet session in the evening with your friends and you forgot to book a place for that in the snack bar. What if you have someone who can interact with people in your place and do the work for you. How comfortable your life will become. Well, yes it is possible with GOOGLE DUPLEX your virtual assistant at work which can make reservations for you, can talk to people on your behalf.

DUPLEX is the AI-driven human-sounding voice which can accomplish a real-world task over phone calls. It can schedule appointments for you,  without any interaction from the user. It calls a restaurant to talk with the person at the restaurant, books a table for you and then tells the original user about the details of the final booking. Moreover, it uses vocabulary fillers while talking which makes it more human-like. It is built to sound as natural as possible. At the heart of Duplex is RNN or recurrent neural network build to counter real-world situations. It uses the output of Google's Automatic Speech Recognition and features from the history of conversation and more.

Google Duplex was first launched in New York, Phoenix, Atlanta, and metropolitan areas of San Fransisco for owners of Google Pixel, Pixel XL, Pixel 2 XL,  Pixel 2, Pixel 3 and Google Pixel 3 XL phones. Recently, Google Duplex support has started in any Android devices running version 5.0 Lollipop or newer. It's also starting Apple's iPhones that have Google Assistant installed. Presently 43 states in the US can use GOOGLE DUPLEX. It can interact with the user as naturally as a human can talk. When talking to a machine, one of the many challenges that pop up is latency. Any delay in replying to the other person can cause communication problems. Google states that RNN takes care of the latency issues. In some cases, they do not even wait for the RNN to react. Instead, they use faster approximations. As of now, Duplex has a response time of around 100ms, which is quite appreciable.

In some cases where complex questions have to be answered, a delayed answer works, just like a normal human being would take time in answering the question after thinking. Adapting to different situations is one of the many strengths of Google Duplex. The technology in Google Duplex involves a self-monitoring feature as well. While Duplex can engage in pretty complex conversations, in cases like handling a complex appointment, Duplex can hand over the task to a human operator. Google Duplex trains through real-time supervised learning. This allows it to understand and evolve in real-life situations just like a human. Real-time learning is stopped when the system reaches the desired level of efficiency and quality.

Google Duplex is changing the way we interact with the world. It saves a lot of time and is a blessing for people with extremely busy schedules. In the near future, Duplex is surely going to become better than it is now. The time is not very far when we would be sleeping while Duplex pays our bills, reserves a table for that dinner, makes our weekend appointment with the doctor and books our tickets for a Friday night movie.