Google Duplex working | What can Google Duplex do
Google Duplex [A brand new Google Assistant]
It’s a behavior of Google assistance. The Google Duplex extends Google assistance that can make phone calls to actual humans just by you asking it to do so. It’s a deep neural network that builds off wavelet technology. The Google Duplex is so that you never have to call a service business ever again.
Google Duplex use Wavenet
The wavenet is a speed synthesis program that is worked by joining very short units of sound together to create speech. It was a breakthrough in natural speech synthesis when it came out. The wavenet is still generating raw audio to randomly imitate human sounds & it still sounds like words.
The Duplex is another neural network built on the top of wavenet. The result is an AI that can have an actual conversation but with wavenet vocal precision. It has trained duplex in the narrow field of scheduling appointments or bookings & inquiring about a business’s opening hours on holidays. The Duplex cannot have a general conversation.
Google stated that the unique challenges for training such a neural net. How do you get an AI to robustly understand natural language & reply realistically? As this will be pretty hard to do. Like–people talk differently to one another than with computers. We talk faster, correct ourselves mid-conversation & even a mid part of conversation & rely on context instead. Though in the poor quality & noisiness of a phone line & you have a pretty hard challenge on your hands.
Working on Google Duplex
Google Duplex uses something known as the recurrent neural network. The neural network basically is a massive matrix multiplication function where each part of the matrix is built up of artificial neurons called nodes. As the nodes contain a mathematical formula arranged in layers. Each node has input & output. After receiving inputs, the end goal of the whole matrix or neural network is basically to find out how to reduce how long it is. As no one actually knows how they come to their conclusion.
You just give them the inputs & they somehow get an answer. The neural network has been around for a while. The specific type of neural network that Duplex is using is called a recurrent neural network. This type of network has a small internal memory that allows them to remember a specific input to help understand the context. For this reason, is the perfect neural network for speech recognition at the heart of most speech recognition algorithms.
How Duplex Trained
The conversation starts with an analog speech. In this speech an audio form fed into Google’s automatic speech recognition system. From this point, the audio now understood as the text & this text once converted into a format of the neural network can understand. Then it is fed into the system. Other metadata & wider context information from the calls such as the correct time for the appointment schedule in question, all the time of day fed into the system. When the neural network listens to a bunch of phone recordings as inputs, it eventually learns how to reduce the amount of error it has. The last train neural network that comes out of all this is Google Duplex.
What can Duplex do
Duplex can also do some pretty interesting things like handling interruptions, elaborate & respond to the audio issues over the phone. The Duplex is not for general conversation, as it is only specifically for booking reservations & inquiring about open hours during holidays. If a store has an online page, it’s far quicker to book that way. A Google Duplex is using a neural network. This technology is pretty cool. Google stated they are going to be transparent when you talking to a duplex AI.