We are all familiar with Gboard, especially those of us using Android. It is a virtual keyboard application released for iOS and Android developed by Google. The first stable version was released in 2016. The application features
- Searching web results through Google search.
- Predictive answering of questions.
- Easy sharing of emoji and GIF.
- Supporting multiple languages and many other features.
What is the Company Saying About This New Feature?
Google’s spokesperson Johan Schalkwyk recently announced this big news in a blog post. He said that the team is rolling out an all neural end to an end speech recognizer that is installed on the device which always works, even when the user is offline.
Google is currently aiming for their keyboard to perform at a character level. Their new model is trained using RNN-T technology. The new RNN Transducer makes the speech recognizer model compact enough to be installed in the device hardware.
Advantages of the Offline Feature?
The major advantage is that the user will not suffer from spottiness or network latency. Since the feature works offline, it works exactly like when someone is typing word by word what you are saying in real time, without any pause or hold. This is what exactly the users need from an inbuilt keyboard dictation system.
To avoid sudden network failure or latency, the new model is hosted on a device, which will increase this speech recognition system’s parameter of usage. For now, it is available in all Pixel devices for only American English.
Any Chance of Multi-Language Support in the Future?
According to Schalkwyk, the all neural speech recognizer of Google that works on the device is primarily released for Pixel devices and only in American English. Based on the trends in the industry and the fact that they are working on improving the algorithm and converging specialized hardware, very soon the developers are hoping that the technology can be adapted to more languages. They are also planning to increase the application’s domain range. For now, let’s hope that very soon, the feature will be released for other Android devices and supports our local dialect as well.