Google has updated Gboard's speech recognizer to mimic the keyboard dictation system where the system recognises characters one-by-one as they are spoken. The system is now directly hosted on the device, eliminating the need to rely on seamless internet connectivity.
As explained by Google on its blog, the new all-neural, on-device Gboard speech recognizer does not search over a larger decoder graph and instead consists of a beam search through a single neural network.
Despite the small size of 80MB, the system is capable of offering the same accuracy as the traditional server-based models.
The ability is currently supported on Pixel devices for American English only.
(This story has not been edited by Business Standard staff and is auto-generated from a syndicated feed.)