Η Google δημοσιεύει εκατοντάδες ερευνητικές εργασίες κάθε χρόνο που καλύπτουν σχεδόν τα πάντα. Κατά καιρούς έχουμε δει πολλούς και διαφορετικούς algorithms, which aim to make our lives easier. So right now a new upcoming project is being "baked" that is getting ready to be presented later this month. It aims to give us the best AI (artificial intelligence) assistant we've ever had.
You may have encountered difficulty when trying to use something like Google Now, and you are not connected to the Web.
So, the Google Research team is trying to develop a new, personalized system for voice commands and dictation that works within your device instead of asking for help from a server. All this without needing too much space on your device.
To train the new project, around 2.000 hours of anonymous Google voice searches, totaling over 100 millions statements, while also using noise from YouTube to mimic noisy real-life conditions.
By implementing a series of complex computational models and plenty of compression in the files to take as little disk space as possible, the team has come up with a local voice system that runs on average seven times faster than real-time on a Nexus 5 while using just 20,3 MB.
As with Google Now voice commands, it can handle the main names and other information from the device to perform special tasks, such as sending an e-mail to a contact. Instructions can be recorded, and saved if you are not online.
The Results of the new project will be presented at 41 International Conference on Acoustics, Speech and Signal Processing, which will start on March 20.